OLLI RAJALA
Oscillator Phase Noise Measurements using the Phase Lock Method
Master of Science Thesis
Examiners: Olli-Pekka Lundén,
Tiiti Kellomäki
Examiners and subject were approved
in the Faculty of Computing and
Electrical Engineering Council
meeting on 13.01.2010
II
ABSTRACT
TAMPERE UNIVERSITY OF TECHNOLOGY
Master’s Degree Program in Signal Processing and Communications Engineering
RAJALA, OLLI: Oscillator Phase Noise Measurements using the Phase Lock Method
Master of Science Thesis, 49 pages, 5 Appendix pages
June 2010
Major: High Frequency Technology
Examiners: Olli-Pekka Lundén, Tiiti Kellomäki
Keywords: measurements, oscillators, phase noise, radio frequency
This thesis covers various methods to measure phase noise of oscillators. The main
goal is to build a working phase noise measurement system for the Department of Electronics, Tampere University of Technology. The current facilities have limitations which
restrict the measurements.
Phase noise of oscillators can be measured for example with the following methods:
phase locked loop (PLL) method, delay line discriminator method, and cross correlation
method. In addition to these basic methods, some commercial phase noise measurement systems will be covered. Phase noise of two-ports can be measured with a residual
method, and this will be covered shortly.
The phase locked loop method is chosen for closer investigations. A measurement
system based on that method is built and all related properties are introduced. Most of
the required components were readily available, but two filters and a low noise amplifier
for 0–100 kHz bandwidth was built and their performance verified during this project.
The design and building process of those components is covered, and the performance of
all the components used is verified with measurements. The measurement process and
set-ups used are introduced.
The measurement system built during this project can be used to measure phase noise
of oscillators, as shown in this thesis. Measuring phase noise of low noise oscillators is not
an easy task, but it can be done without expensive commercial phase noise measurement
systems using only commonly available equipment. This thesis shows how that can be
done.
III
TIIVISTELMÄ
TAMPEREEN TEKNILLINEN YLIOPISTO
Signaalinkäsittelyn ja tietoliikennetekniikan koulutusohjelma
RAJALA, OLLI: Vaihelukkomenetelmän hyödyntäminen oskillaattorien vaihekohi-
namittauksissa
Diplomityö, 49 sivua, 5 liitesivua
Kesäkuu 2010
Pääaine: Suurtaajuustekniikka
Tarkastajat: Olli-Pekka Lundén, Tiiti Kellomäki
Avainsanat: mittaaminen, oskillaattori, radiotekniikka, vaihekohina
Tässä työssä paneudutaan oskillaattorien vaihekohinaan ja sen mittaamiseen käytettyihin menetelmiin. Päällimmäisenä tarkoituksena on kehittää toimiva vaihekohinan mittausjärjestelmä Tampereen teknillisen yliopiston elektroniikan laitokselle. Laitoksen RFLaboratoriossa tällä hetkellä käytetty menetelmä ei mahdollista useimpien oskillaattorien
vaihekohinan mittaamista, koska järjestelmän rajat tulevat nopeasti vastaan.
Oskillaattorien vaihekohinaa voidaan mitata esimerkiksi seuraavilla menetelmillä: vaihelukkomenetelmä (PLL-menetelmä), viivelinjadiskriminaattorimenetelmä ja ristikorrelaatiomenetelmä. Näiden lisäksi käsitellään kaupallisia toteutuksia ja vaihekohinan mittaamista suoraan spektrianalysaattorilla. Kaksiporttien vaihekohina saadaan mitattua esimerkiksi residuaalimenetelmällä, joka esitellään lyhyesti.
Mittausmenetelmistä tarkimmin paneudutaan vaihelukkomenetelmään. Sen ympärille
rakennetun mittausjärjestelmän ominaisuudet esitellään ja kerrotaan eri komponenttien
vaatimuksista. Suurin osa järjestelmän vaatimista komponenteista löytyi laboratoriosta
valmiina, mutta kaksi suodatinta sekä pienikohinainen vahvistin 0–100 kHz taajuuskaistalle täytyi rakentaa. Noiden komponenttien suunnittelu ja rakentaminen käydään läpi riittävällä tarkkuudella. Käytettyjen komponenttien ominaisuudet mitattiin niiltä osin kuin
ne vaikuttavat käytettyyn mittausjärjestelmään. Näistä tuloksista ja mittauksista käydään
läpi järjestelmän suorituskyvylle kriittiset alueet.
Projektin aikana rakennettu mittausjärjestelmä todettiin toimivaksi tavaksi mitata vaihekohinaa. Pienikohinaisten oskillaattorien vaihekohinan mittaaminen on varsin vaativa tehtävä, mutta tämän työn perusteella se on mahdollista tehdä ilman erityisiä kaupallisia vaihekohinamittalaitteita käyttäen hyväksi RF-laboratoriosta todennäkoisesti löytyviä laitteita.
IV
FOREWORDS
This has been a really interesting journey. I have learned much and even though it has
been quite hard once in a while, it has been kind of fun. First I would like to express my
gratitude to the TUT Department of Electronics because they offered me this possibility.
This was not a paid project, but nevertheless, it was still a good thing to have this possibility. Otherwise it would have been necessary to invent my own project, and that is never
easy.
I would like to thank my teachers and examiners Olli-Pekka Lundén and Tiiti Kellomäki for their support and suggestions during this process and all these years. Without
Tiiti’s RF measurement course I would not have seen the light and probably would not
have changed my major. I am also really grateful for Tiiti’s help in Matlab and LATEX related things. Thanks!
Jimmy Turunen has also helped quite a lot in LATEX coding. I know I have had too
many newbie questions, but thanks for all the suggestions and guidance. It would be a
good idea to some day make a LATEX tutorial for all M.Sc. thesis writers, as you suggested.
Thanks for my whole family. Thanks dad for your encouragement and suggestions,
even though the requirements and methods in the technical field are quite different than in
your pedagogics field. It was still interesting to compare our simultaneous thesis writing
projects and it also helped in some level. Thanks mom for your support and thoughts.
And finally, thanks for my lovely wife who has supported and loved me even though
do not understand this field. Thanks also for your wonderful meals and thanks for our
dogs who forced me to go outside so I would not lose my mind. You guys are awesome!
Tampere 27.5.2010
OLLI RAJALA
olli.rajala@ravoltek.net
V
CONTENTS
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Goals of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . .
2. What is Phase Noise? . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3. Measuring Phase Noise . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1 About Phase Noise Measurements . . . . . . . . . . . . . . . . . .
3.2 PLL Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3 Delay Line Discriminator Method . . . . . . . . . . . . . . . . . .
3.4 Cross-Correlation Method . . . . . . . . . . . . . . . . . . . . . .
3.5 Commercial Measurement Systems . . . . . . . . . . . . . . . . .
3.5.1 Measuring Phase Noise with a Spectrum Analyzer . . . . . . .
3.5.2 Phase Noise Measurement Modules of Spectrum Analyzers . .
3.6 Residual Method . . . . . . . . . . . . . . . . . . . . . . . . . . .
4. Building the Measurement Set-up . . . . . . . . . . . . . . . . . . . . .
4.1 Basics of the Measurement Set-Up . . . . . . . . . . . . . . . . . .
4.2 The Low Noise Amplifier and the PLL . . . . . . . . . . . . . . .
4.3 Spectrum Analyzer . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 About different Spectrum Analyzers . . . . . . . . . . . . . .
4.3.2 Problems in using a Laptop Computer as a Spectrum Analyzer
5. Testing the System . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1 Measurement Basics . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Verification of the Measurement Set-up . . . . . . . . . . . . . . .
5.2.1 Agilent E4407B Spectrum Analyzer . . . . . . . . . . . . . .
5.2.2 Mini-Circuits ZX05-42MH Mixer . . . . . . . . . . . . . . .
5.2.3 Low Pass Filter, Attenuators, Cabling and Adapters . . . . . .
5.2.4 Low Noise Amplifier . . . . . . . . . . . . . . . . . . . . . .
5.3 Phase Noise Measurements with the Spectrum Analyzer . . . . . .
5.4 Actual Phase Noise Measurements using the PLL Method . . . . .
5.4.1 Measurement Procedure . . . . . . . . . . . . . . . . . . . . .
5.4.2 Processing the Measurement Results . . . . . . . . . . . . . .
5.4.3 Measuring the free running Oscillators . . . . . . . . . . . . .
5.5 Error Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.1 Goals of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.LNA Schematics and PCB Layout . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
1
2
2
4
7
7
8
10
12
13
14
15
16
17
17
18
22
24
24
25
25
26
27
28
31
33
34
36
37
39
44
45
47
47
47
50
53
VI
LIST OF FIGURES
2.1
2.2
A general phase noise spectrum . . . . . . . . . . . . . . . . . . . . . .
The different areas of a phase noise frequency plot . . . . . . . . . . . .
3.1
3.2
3.3
3.4
3.5
3.6
3.7
PLL method set-up . . . . . . . . . . . . . . . . . . .
The expected phase noise measurement floor . . . . .
Delay line discriminator set-up . . . . . . . . . . . . .
Cross correlation method set-up . . . . . . . . . . . .
Spectrum analyzer method set-up . . . . . . . . . . . .
What determines the resolution of a spectrum analyzer
Residual method set-up . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
10
11
12
14
15
16
4.1
4.2
4.3
4.4
PLL method set-up . . . . . . . . .
Amplifier block diagrams . . . . . .
The PCB of the low noise amplifier .
Switch settings . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
18
19
22
22
5.1
5.2
5.3
The published phase noise performance of the ESG-D4000A . . . . . . .
Measured phase noise when using different PLL modes of the ESG-D3000A
Phase noise of different oscillators, measured with the phase noise module
of E4407B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Phase noise specifications of Agilent E4407B . . . . . . . . . . . . . . .
The measurement set-up used to measure the mixer DC level . . . . . . .
Normalized mixer output voltage vs. phase difference of inputs . . . . . .
The schematics of the Butterworth low pass filter . . . . . . . . . . . . .
Measurement set-up used for measuring the (amplitude) response of the
low pass filter, the attenuators and the LNA . . . . . . . . . . . . . . . .
Measured frequency response of the low pass filter, −3 dB point at 4.2 MHz
The frequency response of the amplifier . . . . . . . . . . . . . . . . . .
The variation in gain compared to the mean gain in the shown bandwidth
Phase noise of the oscillator B measured three times with the E4470B . .
Phase noise of the oscillator B measured five times with the KE5FX GPIB
toolkit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Phase noise measured with the spot frequency mode . . . . . . . . . . . .
The schematics of the final measurement set-up. . . . . . . . . . . . . . .
The final measurement set-up. . . . . . . . . . . . . . . . . . . . . . . .
The beat note and measured noise level as seen on the screen of the spectrum analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Difference in the measured phase noise when using the different DC levels
5.4
5.5
5.6
5.7
5.8
5.9
5.10
5.11
5.12
5.13
5.14
5.15
5.16
5.17
5.18
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5
6
26
27
28
29
29
30
31
31
32
34
35
36
37
38
39
40
41
42
VII
5.19 Repeatability of the measurement . . . . . . . . . . . . . . . . . . . . .
5.20 Phase noise of the oscillators A, B and C . . . . . . . . . . . . . . . . . .
43
44
6.1
Two variations of the same measurement set-up . . . . . . . . . . . . . .
48
A.1 The schematics of the low noise amplifier . . . . . . . . . . . . . . . . .
54
VIII
TERMS AND SYMBOLS
Symbols
A
B
c
E
f
f0
fm
F
G
H
k
LSSB
l
φ0
P
QL
S
σ
t
T
V
VCC
vf
Vtune
ω0
U
Z0
Amplitude of a signal
Equivalent Noise Bandwidth
Speed of light
Error
Frequency
Oscillator carrier frequency (Hertz)
Frequency offset from carrier (Hertz)
Device noise factor at operating power level A
Gain
A correction factor
Boltzmann’s constant, 1.38 ∗ 10−23 Joule/Kelvin
Single-sideband phase noise density (dBc/Hz)
Length
Phase of a signal
Signal power
loaded Q (dimensionless)
Shape factor of the filter
Standard deviation
Time
Temperature
Signal
Supply voltage
Velocity factor
Tuning voltage of a VCO
Angular frequency of a signal
Voltage
Characteristic impedance
IX
Terms
AC
AM
ATT
BNC
dBc
dBm
DC
DPDT
DUT
FET
FFT
FM
FR4
GPIB
IF
LNA
LO
PCB
PLL
PM
RBW
RMS
RSS
RF
SMA
SMD
SSB
USB
TUT
VCO
Alternating current
Amplitude modulation
Attenuator
Bayonet Neill-Concelman -connector
Level relative to the level of the carrier, in decibel scale
Power relative to 1 mW, in decibel scale, calculated as 10 log10
Direct current
Double pole double throw -switch
Device under test
Field effect transistor
Fast Fourier transform
Frequency modulation
Flame resistant PCB material
General purpose interface bus
Intermediate frequency
Low noise amplifier
Local oscillator
Printed circuit board
Phase locked loop
Phase modulation
Resolution bandwidth
Root mean square
Root sum square
Radio frequency
Subminiature version A -connector
Surface mount device
Single side band
Universal serial bus
Tampere University of Technology
Voltage controlled oscillator
Power
1 mW
1
1.
INTRODUCTION
Measuring physical phenomena accurately is both important for the science and difficult
to do with high resolution. This thesis tries to give some information about one particular
phenomenon, phase noise of oscillators, and how that can be measured with the required
accuracy.
The first chapter covers the general things about this thesis. There will be some
thoughts about the importance of this topic, what are the goals of this thesis and how
this thesis is structured. The main motivation behind this thesis is to help practical every
day measurement procedures in an RF laboratory and thus all theory and measurements
will be presented with that in mind.
1.1
Motivation
Noise is everywhere. Sometimes it is wanted, usually not. Broadband noise is a good
test signal when measuring for example the frequency response of an amplifier, because
it includes all frequencies of interest. On the other hand, in communication systems
noise is not a wanted phenomenon, because it disturbs the used signal and thus makes
communication harder. It can thus be said that minimizing noise is a very important thing
in every communication system. It is not possible to totally eliminate noise, but there are
many ways to reduce it. When the used modulation schemes, circuits, and components
are designed correctly, noise properties and noise immunity in a particular system can be
significantly improved. During the design and evaluation process it is important to use
the correct methods to measure all the necessary properties.
Measuring amplitude and frequency related properties is often a straightforward process. Phase noise, on the other hand, is more difficult to measure, still it is an important
property when designing communication systems with good selectivity and sensitivity.
Especially the phase noise of the oscillator(s) in receivers plays a very important role in
selectivity.
A receiver adjacent channel selectivity, the total signal-to-noise ratio of a receiver and
the velocity resolution of a Doppler radar are among things which are affected by single
sideband (SSB) phase noise. [1] Without a method to measure phase noise accurately, it is
impossible to understand all limitations in the communication system. This understanding
is important when a designer tries to improve the communication system.
1. Introduction
1.2
2
Goals of the Thesis
The main goal of this thesis is to build a phase noise measurement system for the radio frequency (RF) laboratory of the Department of Electronics, Tampere University of
Technology (RF-Laboratory in the following sections). Different phase noise measurement systems are also investigated during this project even though only one is built and
its performance verified. The emphasis is on radio frequency oscillator measurements,
but naturally most of the procedures described in this thesis can be adapted to measuring other oscillators as well. After reading this thesis, the reader should be familiar with
phase noise measurements and related problems which will arise when building a suitable
measurement set-up.
It was known in advance that the current phase noise measurement system in the RFLaboratory was not adequate. At the moment phase noise is usually measured directly
with a spectrum analyzer, and this method is a limited one. Basically it is possible to
measure only oscillators in a limited performance range, because the system used sets
some limitations for the performance of the oscillator. This method can not be used to
measure oscillators with good phase noise performance, because the noise floor of the
measurement is too high. Also oscillators which drift much are problematic, because the
carrier frequency should be steady during the measurement. This particular method will
be covered and the problems discussed in more detail in sections 3.5.1, 3.5.1, and 5.2.1.
At the beginning of this project it was planned that a working measurement set-up
would result from this project. During the project that goal changed somewhat, and finally
the goal was to acquire as much information of phase noise measurements as possible.
If some working solution could be found, necessary components would be ordered or
built and the measurement set-up assembled. Finally the performance of the assembled
measurement set-up would be verified thoroughly.
1.3
Structure of the Thesis
Chapter 2 describes the theory behind phase noise. The discussion is limited to the parts
necessary for this thesis. References to more detailed sources are given.
Chapter 3 reviews the different measurement set-ups and also introduces the problems
found in those set-ups. It is necessary to know the weaknesses and strengths of different
methods to choose a suitable one.
Chapter 4 describes in more details the chosen measurement set-up, which is called
the PLL method. Different parts of the set-up are introduced and important properties
discussed. This chapter covers also some specific things which must be considered when
building this or any other measurement system.
Chapter 5 shows the measurement results and provides more information on how the
actual measurements were done. The chapter begins by describing all necessary prepar-
1. Introduction
3
ative measurements and continues finally to the results from the actual phase noise measurements. Problems arising during the different phases of the measurement process will
be covered and solutions suggested, if possible. One section is also reserved to error
analysis.
Chapter 6 discusses on the biggest weaknesses of the chosen measurement set-up and
gives some ideas how to improve the performance. There are some areas where more
research is needed, and that information can be found in this chapter. This chapter also
summarizes the whole project.
4
2.
WHAT IS PHASE NOISE?
Phase noise is closely related to the performance of oscillators. Naturally the phase noise
of a whole complex system can be measured, but it can be closely approximated by adding
the phase noises of different oscillators together. In this context it is sufficient to define
phase noise and all the related properties through the behavior of an oscillator.
The output of an ideal oscillator can be expressed as
Vout (t) = Aout sin (ω0t + φ0 )
(2.1)
where Aout is the amplitude, ω0 the angular frequency and φ0 the phase angle of the
signal. Eq. (2.1) describes a theoretical situation because it does not take any kind of
noise into account. When also noise is considered, the amplitude and the phase angle
become dependent of time and the equation can be written as: [2, 3]
Vout (t) = Aout (t) sin [ω0t + φ (t)]
(2.2)
In a well designed and behaving oscillator the amplitude is so stable that it can usually be
considered constant. Usually the noise level caused by the amplitude modulation component (Aout (t) in the equation) is 20 dB lower than the noise level caused by the phase
noise component (φ (t) in the equation), because most oscillators operate in saturation.
Thus phase noise is the main noise contributor in an oscillator. [2]
The phase can vary either discretely or continuously. The previous one produces discrete signals called spurious frequencies, which can be seen as discrete components in the
spectral density plot. The continuous part of the phase fluctuation causes the phenomenon
called phase noise. It is a random phenomenon and thus can only be described with statistical terms. It produces sidebands to the spectral density plot which can be observed
for example with a spectrum analyzer. Figure 2.1 show an example plot which includes
phase noise and some spurious signals. [2]
Phase noise has been studied much in the past, and there are many equations covering
its behavior. Probably the most used one these days is the Leeson’s equation [1, p. 230]
"
LSSB = 10 log10
FkT 1
P 8Q2L
f0
fm
2 #
(2.3)
2. What is Phase Noise?
5
Figure 2.1: A general phase noise spectrum
where
LSSB = single-sideband phase noise density (dBc/Hz)
F = device noise factor at operating power level A (dimensionless)
k = Boltzmann’s constant, 1.38 × 10−23 J/K
T = Temperature (K)
P = oscillator output power (W)
QL = loaded Q (dimensionless)
f0 = oscillator carrier frequency (Hz)
fm = frequency offset from carrier (Hz)
Eq. (2.3) applies when fm is greater than the f1 = 1/ f -flicker corner frequency of the
active device and smaller than f2 = f0 /2QL . In (2.3), f1 is typically less than 1 kHz and
f2 is the frequency where white noise level starts to dominate, usually some MHz. This
is shown more clearly in Figure 2.2. There are also some other restrictions which must
be met before the equation can be used: F must be known; loaded Q must include all
effects from component losses, and device and output buffer loading; and the oscillator
must include only a single resonator. [1]
Leeson’s equation identifies the different parameters affecting the phase noise performance in an oscillator circuit. It can also be easily seen what parameters must be tuned
and how if we want to reduce phase noise. Some of them can be easily measured or de-
2. What is Phase Noise?
6
9 dB / octave
6 dB / octave
flat
f1
f2
Figure 2.2: The different areas of a phase noise frequency plot
termined, but some are more problematic. The most complicated one is the noise factor
of the oscillator (F), because of the non-linear operation of the active devices in the oscillator circuit. In practice F is often not measured directly. Loaded Q is obtained with a
suitable technique, and after measuring SSB phase noise, F is calculated from Leeson’s
equation. [1] So, even though Leeson’s equation can be used to calculate the phase noise
of an oscillator, it is usually quite difficult to use it for that, but during the design process
it can be used as a guide and a checking tool.
Phase noise is usually expressed as dBc/Hz at some specific offset from the carrier.
That means noise level relative to the carrier level and calculated in 1 Hz bandwidth.
Often only single sideband noise is considered, because the noise sidebands are assumed
to be approximately symmetric around the carrier. Many measurement set-ups measure
both sidebands and the final result is an average between the sidebands.
Phase noise describes how the frequency of an oscillator varies in short time scale.
The long term frequency stability is called frequency drift, and it must also be considered
during the measurement process. The output frequency of an oscillator takes some time
to stabilize after the oscillator has been started and this drift can be up to dozens of MHz
as was noticed during this project. The output frequency also usually drifts noticeably
during the measurements, especially in the case of free running oscillators. This drift is
a real problem, because during the measurements the system must be able to lock to the
carrier or the carrier must be stable enough. If a carrier tracking mechanism is used, the
drift is not such a big problem anymore.
7
3.
MEASURING PHASE NOISE
Usually there are more than one way to measure the phenomenon of interest. This is
a good thing, because all measurement systems have strengths and weaknesses. The
engineer must understand the limitation so a suitable method can be chosen.
This chapter covers the different phase noise measurement methods and some space
is also reserved for the discussion about the general things related to all measurements.
The phase noise measurement methods covered in this thesis are PLL method, Delay
line discriminator method, Cross-correlation method, and Commercial products. The
first one is the simplest and has the biggest limitations, and the second last one requires
the most complex measurement system, but is also the most versatile one. The Crosscorrelation method can even measure oscillators with a better phase noise performance
than the reference oscillator.
3.1
About Phase Noise Measurements
Ray Kammer said when he was the director of National Institute of Standards and Technology: ’If I have one watch, I always know what time it is. If I have two watches, I’m
never quite sure what time it is’. [4] This is the basic problem in all measurements: getting
reliable and reproducible results is difficult. Especially when measuring some non-static
property, such as a real world signal, deviations in results are unavoidable. These deviations depend mostly on the measurement accuracy. If 1.5 GHz is accurate enough
frequency for an oscillator, it is usually possible to get that accurate result all the time,
but if the required accuracy is 1.500 000 000 GHz, some deviations will be seen in the
results. The measurement resolution is thus one of the things which must be considered
when choosing a suitable measurement system.
In general, measuring phase noise is more difficult than measuring amplitude or frequency related properties. In reality this of course depends on the required accuracy. The
biggest reason for this is the huge dynamic range required in the phase noise measurement process. Good oscillators can have phase noise in the range of −120 dBc/Hz at a
10 kHz offset from the carrier. Another limiting property is the phase noise of the measurement instrument, which can easily prevent using some of the techniques described in
the following sections.
There are many different phase noise measurement techniques, and the right one must
be chosen when performing measurements. It is necessary to know the weaknesses and
3. Measuring Phase Noise
8
strengths of different systems, because none of these methods is a perfect choice for every situation. Often the most limiting part is still the availability of measurement systems,
because specialized phase noise measurement equipment is relatively expensive. Fortunately there are some in-expensive systems, and one of them is built and its performance
investigated during this project. Section 3.2 describes in more details the chosen method,
which is called the PLL method in this thesis.
There are many ways to categorize phase noise measurement systems. One way is to
divide them based on how the signal is measured, into direct method and in-direct methods
with some form of a conversion. In the first ones phase noise is measured directly from
the signal and in the latter ones the signal is somehow converted to a form which is
more suitable for measurement purposes. Usually the in-direct method does not directly
measure the phase noise, but some other property, for example the error signal in a PLL
loop, is measured and phase noise is calculated from the results.
3.2
PLL Method
Because many measurements are difficult to make in high frequencies and much easier
near 0 Hz, it is often easier to bring the output of the DUT to near 0 Hz, if that will not
alter the interesting property too much. Every traditional radio transmitter and receiver
makes this conversion, so components for that conversion can be found easily. Naturally
quality is a much more important feature in measurement systems than for example in
consumer radio receivers, so care must be taken when choosing the right way to do this
conversion.
In [5, p. 9.10] a low cost method for measuring phase noise is introduced. The measurement set-up can be seen in Figure 3.1. In that method there is a reference oscillator,
which should have as low phase noise as possible. The phase noise of this oscillator is
the most limiting factor. The signal of the reference oscillator is mixed with the signal of
the DUT and the reference oscillator is kept locked to the DUT with a phase locked loop
(PLL).
PLL is basically a system that compares the phases of two input signals and generates
a third signal which can be used to steer one of the input signals closer to another. When
the phases of the input signals are aligned, the loop is said to be locked and the control
signal is zero. The mixer behaves as a phase detector and its output is zero when the input
signals are in quadrature. References [6] and [7] describe in detail why a mixer can be
used as a phase detector. In this thesis, the varying DC level is the control signal which is
used for locking the PLL.
When the frequencies of the input signals are close enough to each other, the output
of the double-balanced diode mixer is basically a DC voltage which relates to the phase
difference between the input signals. This voltage varies a little due to phase noise of the
input signals. It should be remembered that the noise present at the output of the mixer
3. Measuring Phase Noise
9
Signal
Generator
DUT
Low
Pass
FFT/
Spectrum
Analyzer
DC
Block
Measured
Signal
Control signal
Oscilloscope
Figure 3.1: PLL method set-up
includes phase noises of both oscillators. If the noise from the reference oscillator is more
than 20 dB lower than the noise from the DUT, the main contributor for phase noise is
the DUT, as Table 3.1 shows. This topic will be discussed later in this section. The little
voltage variations in the PLL control voltage are amplified and measured with a spectrum
analyzer. The method for measuring the frequency content of the signal is not critical.
For example an FFT or a traditional RF spectrum analyzer can be used. The exact DC
level of the signal is not interesting, but the interesting part is the variation in the control
voltage. The DC level is thus filtered away with a DC block. [6], [5, p. 9.12]
This method is relatively easy to deploy and can be used to measure phase noise down
to 170 dBc/Hz at 10 kHz offset, as Figure 3.2 shows. It also shows noise floors of two
other measurement techniques, which will be described in the next sections. [8]
The PLL method is also suitable for very broadband measurements. With only a few
different mixers and suitable reference oscillators, frequencies from 1 MHz to dozens of
GHz can be measured. The main problem with this method is that it is not possible to
know which part of the noise comes from the reference and which from the DUT, but this
is the problem in most measurement systems. If the used reference is good enough, so
that its noise is 20 dB lower than the noise of DUT, it can be neglected. If the noise levels
are not that far from each other, a correction factor must be subtracted from the result.
The correction factor is from 0 to 3 dB, where the highest number is used when the noise
levels are equal. [8]
Table 3.1 shows the correction factors for different noise level differences. They can
be calculated as
−∆P
10
Pcorr = 10 log10 1 + 10
dB
(3.1)
where Pcorr is the number that must be subtracted from the results, and ∆P is the difference
between the noises of the reference and the DUT in dB. [9]
3. Measuring Phase Noise
10
−20
Delay discriminator, 500 ns delay
PLL
Cross−correlation
−40
−60
dBc/Hz
−80
−100
−120
−140
−160
−180
−200
1
10
100
1k
10k
100k
Frequency / Hz
Figure 3.2: The expected phase noise measurement floor
Table 3.1: The correction factors if the phase noise of the reference oscillator is near the phase
noise of the DUT
∆P/dB
Pcorr /dB
0
3
2
2.12
4
1.46
6
0.97
8
0.64
10
0.4
15
0.14
20
0.04
If the DUT is a VCO this method can be altered a little. The reference oscillator will
then be a free running one, and the DUT would be controlled with the PLL. We just need
a suitable PLL amplifier after the low pass filter. This variation and a suitable amplifier is
described in [10] and [11]. The altered version can measure only VCOs, which must be
considered when choosing the measurement set-up.
The method described in this section was used in this thesis even though the real measurement set-up is more complex. The configuration of the used measurement system is
described in more details in Chapter 4 and its performance is verified in Chapter 5.
3.3
Delay Line Discriminator Method
Like the PLL method, the delay line discriminator method utilizes a mixer, but the methods are otherwise not similar, as can be seen in Figure 3.3. For example no reference oscillator is needed. In this method the output of the DUT is amplified and then ran through
a 10 dB directional coupler. The through path of the coupler is connected to a delay line,
3. Measuring Phase Noise
11
Delay
Line
DUT
Directional
Coupler
Phase
Shifter
Low
Pass
FFT/
Spectrum
Analyzer
DC
Block
Figure 3.3: Delay line discriminator set-up
for example 100 m of cable, and then to the mixer input. The 100 m cable has been used
in [8], because it provides around 480 ns of delay, as Eq. (3.2) shows, and usually around
10 dB of attenuation. Thus the amplitude of the delayed signal is about the same as the
amplitude of the coupled signal which was attenuated 10 dB in the directional coupler.
[8] If a delay line without any additional attenuation is used, the directional coupler can
be changed to a power splitter.
The following equations shows how the delay from the specific cable can be calculated.
The equation offers a theoretical value for the delay, which should be close to the real one,
but it is not exactly the same. Connectors add delay and the cable properties vary inside
the cable.
tdelay =
lcable 100 m
=
≈ 480 ns
vf c
0.7c
√
vf = 1/ εr
(3.2)
(3.3)
where vf is the velocity factor and εr is the relative dielectric constant of the dielectric in a
coaxial cable. εr is usually from 1.3 to 2.3 and thus vf is in the range of 0.66 to 0.88 [12].
The coupled signal is connected to a phase shifter and then to the other input of the
mixer. The phase shifter is used to adjust the mixer inputs to quadrature. Again the output
of the mixer is measured. This time the output is not directly the phase noise, but it
must be converted to the right form. When the mixer sensitivity is calibrated correctly,
the results must be divided by f 2 to get phase noise level, where f is the used offset
frequency. The mixer sensitivity can be obtained by shifting the carrier frequency up and
down from the center frequency to get approximately ±1 V change. [8]
The good thing with this method is that it can be used to measure noisy sources. On
the other hand, it does not work with good sources, because the noise performance of this
method is the limiting factor, as Figure 3.2 shows. The noise floor depends on the length
3. Measuring Phase Noise
12
Reference
oscillator 1
DC
Block
Power
Splitter
FFT
DUT
DC
Block
Reference
oscillator 2
Figure 3.4: Cross correlation method set-up
of the delay. The longer the delay the lower the noise floor, but it will also mean higher
losses and lower offset frequency. [8]
The highest usable offset frequency depends mostly on the length of the delay. There is
1
a null at f = tdelay
offset frequency, and the recommendation is to use offset frequencies up
to f =
[8]
3.4
1
4tdelay .
With a 500 ns delay, the usable offset frequency range is from 0 to 500 kHz.
Cross-Correlation Method
The cross-correlation method is built around a similar measurement set-up as the PLL
method, described in Section 3.2, although it is more complex. As can be seen in Figure 3.4, there are two reference oscillators, one power splitter, two mixer/amplifier/PLL
circuits and a cross-correlation FFT analyzer. The output of the DUT is connected through
a power splitter to two mixer circuits where it is mixed with the signal from those two reference oscillators. The outputs of the mixer circuits are used for PLL circuits as in the
PLL method and these PLL signals are again of interest. The mixer output signals are
then amplified, the DC is filtered away and finally the signals are fed to two channels
of the FFT analyzer. The FFT must be able to perform a cross correlation measurement
between the two input signals. This results in preserving the common noise, whereas the
noise which is not common in both signals is attenuated. The common noise is the noise
from the DUT because it is the same in both signals and the noise which is not common
comes from the two reference oscillators. [8]
3. Measuring Phase Noise
13
This method offers 15 to 20 dB better noise performance than the PLL method, so it
can be used to measure oscillators with lower phase noise. It is even possible to measure
oscillators with better noise performance than the reference oscillators, because the noises
from the reference oscillators are reduced significantly. [8]
On the downside, wit the cross-correlation method, many measurements must be made
and the average calculated between them. Thus the measurement takes longer, and the
DUT must be kept locked longer time. One sweep takes approximately 10 s, and the
required amount of sweep is 2n where n > 2. With a noisy source this may not be easy.
[8] During the measurement phase of this project, it was noticed that even three sweeps
was too much when using the PLL method if the DUT was a free running oscillator.
The frequency of the DUT drifted so much that the PLL went out of lock during the
measurement. By carefully tuning the frequency of the reference oscillator the lock could
be preserved. This is probably a limitation in the used PLL scheme, but no alternative
methods were available for testing.
The cross-correlation method is thus suitable only for measuring good oscillators with
small drift. It also requires more equipment when compared to the PLL method, though
the required components are easy to find. Often there is a power splitter available in an RF
laboratory, and suitable mixers are not very expensive. For example Mini-Circuits ZX0542MH-S+ costs around 30 A
C at the time of writing this thesis and probably the price is
not a problem. In this thesis the cross-correlation method was not used, but in the PLL
method the mixer was ZX05-42MH. It is the same model as ZX05-42MH-S+, except a
bit newer generation. The mixers and low noise amplifiers should be the same model, so
that their performance would be similar.
3.5
Commercial Measurement Systems
Many companies manufacture specialized phase noise measurement systems, so it is not
necessary to assemble one in the laboratory. These systems are often easy to use, and
their performance has been optimized and it is known, so oscillators with really low phase
noise can be measured. The downside is that they are not in-expensive. Even when using
these specialized systems, it is important to understand the underlying phenomena and the
limitations in the system. Unfortunately not all manufacturers describe how their system
works. This section will introduce some commercial products, and the used measurement
method is given if it has been published.
Wenzel Associates offers three different systems: BP-1000-SC (PLL method), BP1000-SC (Cross-correlation), BP-1000-RM (Residual) [13]. With these devices carrier
frequencies from 5 MHz to 1.5 GHz with offsets up to 100 kHz can be measured with the
noise floor down to −190 dBc/Hz.
Agilent Technologies currently offers E5505A Phase Noise Measurement Solution. It
is a modular system designed for phase noise measurements, which can be configured to
3. Measuring Phase Noise
14
DUT
Spectrum
analyzer
Figure 3.5: Spectrum analyzer method set-up
fulfill the specific needs of a customer. With additional modules the system can measure
carrier frequencies from 50 kHz to 110 GHz with offsets from 0.01 Hz to 100 MHz. The
noise floor is around −180 dBc/Hz.
3.5.1
Measuring Phase Noise with a Spectrum Analyzer
Phase noise can even be measured directly with a spectrum analyzer. This is not a very
good alternative if one wants to measure phase noise close to the carrier frequency, mostly
because spectrum analyzers have their own noise properties which can degrade the measurement results. Even in many real life cases the phase noise of DUT is lower than
the noise of the spectrum analyzer and thus it becomes impossible to measure the phase
noise of the DUT. This method does not require complex outboard measurement set-up,
as Figure 3.5 shows. [14, p. 162]
In addition to the limitations in phase noise performance of spectrum analyzers, drifting makes the measurements more problematic. As previously stated, phase noise is
measured relative to the carrier and thus it is important to lock the measurement system
to the carrier, no matter how much the carrier frequency varies due to noise. In practice
the DUT must have a low enough frequency drift when compared to the sweep time of
the spectrum analyzer. Otherwise it is impossible to measure phase noise, because the
carrier frequency varies too much. In general, synthesized sources have a low enough
drift that they can be measured, if the reference oscillator has low enough phase noise.
Free running oscillators, on the other hand, usually drift too much to be measured using a
spectrum analyzer. [14, p. 162]
Take for example a good 10 GHz cavity oscillator which will drift typically 30 kHz/ min.
It takes typically at least 20 s to scan a 50 kHz span with 100 Hz resolution bandwidth.
During that time the frequency of the DUT will have drifted 10 kHz. It is thus easy to
see that this method has some big limitations. [3] This example was from the 1980s and
spectrum analyzers have been improved much since then, but the underlying problem is
the same. With the Agilent E4407B spectrum analyzer the same sweep takes 4 s, so the
frequency of the DUT would have drifted only 2.5 kHz.
As can be seen from the previous example, with the direct spectrum analyzer measurements care must be taken when deciding the right values for the measurement parameters.
In addition, if the resolution bandwidth is too large, the carrier seems to be too wide,
its level too high at the envelope detector at offset frequencies, and the measurement re-
3. Measuring Phase Noise
15
Figure 3.6: What determines the resolution of a spectrum analyzer (Published with permission) [3]
sults are thus misleading. [14, p. 165] Figure 3.6 shows some of those properties which
determine the resolution of a spectrum analyzer.
The frequency resolution of a spectrum analyzer depends mostly on the intermediate
frequency (IF) filter and the stability of the local oscillator (LO). The resolution bandwidth
(RBW) depends thus on the IF filter characteristics. More information about this and other
related parameters can be found at [15].
3.5.2
Phase Noise Measurement Modules of Spectrum Analyzers
Some spectrum analyzers are equipped with methods for calculating phase noise during
the measurement process. Usually these must be bought separately and in the past they
have been expensive. This kind of a phase noise kit is included in the Agilent E4407B
spectrum analyzer in the RF-Laboratory. It can measure phase noise directly, or with
a spot-frequency type of measurement. In the latter one the analyzer tracks the carrier
frequency, and compares its level to the noise level in one offset frequency. This can be
used when the carrier drifts too much for the direct measurement, but the biggest problem
is that this method is really slow. All interesting offset frequencies must be measured
individually and thus the more resolution is wanted to the phase noise plot the slower the
measurement process becomes. It takes about 30 seconds to measure one offset frequency.
If there are too few measured frequencies, some anomalies in the phase noise plot can be
left unnoticed, and with too many frequencies the process takes too long.
3. Measuring Phase Noise
16
DUT
Reference
oscillator
DC
Block
Power
Splitter
FFT
Phase
Shifter
Figure 3.7: Residual method set-up
One alternative for the proprietary phase noise measurement options of spectrum analyzers is a KE5FX GPIB Toolkit written by a radio amateur John Miles, KE5FX. [16]
The program runs on Windows, and it can use dozens of different spectrum analyzers controlled through a GPIB bus for making the actual measurements. The more information
about this can be found in Chapter 5.
To summarize, measuring phase noise directly with a spectrum analyzer is usually not
adequate. It can be used if the performance of the measured oscillator falls in a specific
range, but more often than not this is not the case. If the drift of the oscillator is small
enough to make the measurement possible, its phase noise is typically so low that it is
below the phase noise of the spectrum analyzer and thus the measurements can not be
made. Many components in spectrum analyzers add their own phase noise, for example
local oscillators, different amplifiers, and mixers. These noise contributions add to each
other when performing measurements, and the phase noise performance of a spectrum
analyzer includes all of these. On the other hand, if the phase noise of the DUT is so
high that measurements could be made, the frequency is usually not stable enough for the
measurements to be performed directly with a spectrum analyzer.
3.6
Residual Method
The methods shown thus far can be used to measure only oscillators. There are some
methods for measuring two-port devices, and the residual method is one of them. It can
be used for example to measure amplifiers, mixers, cables, and filters. Figure 3.7 shows
the measurement set-up. In this method the output of a reference source is split with
a power splitter. One branch is connected through the DUT to the mixer and the other
branch through a phase shifter to the mixer. The phase shifter is adjusted until the phases
are in quadrature, and the output of the mixer is measured with a spectrum analyzer.
Because the noise from the reference source is coherent at the mixer input and the signals
are in the quadrature, it will mostly be canceled out. The remaining phase noise at the
mixer output is thus added by the DUT. [17]
17
4.
BUILDING THE MEASUREMENT SET-UP
Time spent planning all the details and parameters applying for the measurement set-up
is time well spent. It is important to know the limitations of all the components so that the
effect of those limitations can be minimized. It is necessary to check the bandwidth and
power specifications because otherwise the signal under investigation may be distorted
too much inside the set-up or it may even break something.
This chapter describes the used measurement set-up in more detail. First there is a
general description about it, how it works and what kind of parts are required. After
that different parts are analyzed more thoroughly and restrictions and problems related
to all parts are discussed. After this chapter the reader should be familiar what really is
required when building a phase noise measurement system using the PLL method used in
this thesis.
4.1
Basics of the Measurement Set-Up
Although the measurement system based on the PLL method is not a really complex one,
there are many things which must be considered. The set-up utilizes some commercial offthe-shelf components and the rest were built during this project. Some of the components
require a more detailed analysis, and some can be covered with fewer words, but all
important parameters are discussed.
As Figure 4.1 shows, the setup consists of two oscillators, a mixer, a low pass filter,
and a low noise amplifier. The output of the circuit is measured using a spectrum analyzer.
The oscilloscope helps by showing when the system is in phase-lock, but it is not required.
Discussion about the different components and their important properties will follow next.
Reference oscillator: The reference oscillator is probably the most important part of
the system. With the PLL method it is possible to measure oscillators with equal or worse
phase noise performance than the reference oscillator. If the phase noise performance of
the reference oscillator is not good enough, its noise can not be separated from the noise
of the DUT. In practice, if the difference is more than 20 dB, the noise of the reference
oscillator can be neglected. As was discussed in Section 3.2, if the noise levels are close
to each other, a correction factor of 0 to 3 dB must be subtracted from the results. This
correction factor is described in Eq. (3.1) and Table 3.1. The reference oscillator should
also be stable, so that it would not drift too much during the measurements.
4. Building the Measurement Set-up
18
Signal
Generator
DUT
Low
Pass
FFT/
Spectrum
Analyzer
DC
Block
Measured
Signal
Control signal
Oscilloscope
Figure 4.1: PLL method set-up, same as Fig. 3.1, shown here for convenience
Mixer: A basic double balanced diode mixer can be used, so this part is not so critical.
The main requirement for the mixer is that it must have a DC coupled port. Double
balanced mixers have often better isolation characteristics and the DC offset is also often
better than with signal balanced mixers. There are also fewer undesired frequencies at the
output of the double balanced mixer than at the output of the signal balanced one. [6]
Low pass filter: The low pass filter following the mixer is also not critical, because the
frequency of the signal to be measured is so low when compared to the RF frequencies
which will be blocked. The performance of the mixer and low pass filter must be known,
but it is not necessary to use the latest innovations in their designs.
Low noise amplifier: The low noise amplifier, in turn, is a critical component that
must be designed and built well. With careful planning its noise properties are not the
limiting part of this measurement set-up. No suitable amplifier was easily available, so
one was built during this project. The design and building process is described more
thoroughly in the following section.
In this project, the Agilent 8648C signal generator was used as the reference oscillator,
and Mini-Circuits ZX05-42MH as the mixer. An analog 40 MHz oscilloscope (Kenwood
CS-5135) was used as a help in the PLL tuning.
4.2
The Low Noise Amplifier and the PLL
The low noise amplifier (LNA) is used to amplify the weak ripple voltage in the PLL
circuit. The bandwidth requirement is from 0 Hz to some hundreds of kHz and it is the
most limiting part in this measurement set-up. The bandwidth of the LNA defines what
is the highest offset frequency that can be actually measured. In this project the upper
−3 dB corner frequency was specified to be between 200 kHz and 300 kHz. The LNA is
located right after the mixer and the low pass filter.
4. Building the Measurement Set-up
19
Different capacitor sizes
for different high pass
frequencies
G = 30 dB
Low pass
section
G = 0 dB
G = 30 dB
G = 0 dB
PLL Amp
(a) Original amplifier
G = 30 dB
Low pass
section
G = 0 dB
G = 30 dB
G = 0 dB
(b) Modified amplifier
Figure 4.2: Amplifier block diagrams
Wenzel Associates Inc. manufactures phase noise measurement systems, and they have
provided schematics for a low noise amplifier targeted for this use [10]. The measurement
system used in this thesis is somewhat different from the system specified in the document
provided by Wenzel Associates, so it was necessary to modify the amplifier circuit to
fulfill these needs. Figure 4.2a shows the original block diagram, and Figure 4.2b the
modified one.
The block diagrams are almost the same, with only two differences. The modified one
has only one high pass option when using the 60 dB option and the PLL amplifier section
is also removed. It was decided to drop all unnecessary components, and one high pass
option was enough for verifying that this amplifier and whole measurement set-up would
work. The PLL amplifier is not required in the set-up used in this thesis, because the signal
generator can be tuned with the signal from the main output of the LNA. If the signal
generator had been replaced by a voltage controlled oscillator (VCO), the PLL amplifier
would have been necessary. In that case the output of the LNA is not adequate, because
usually the tuning voltage is 0 to 5 V or something similar. In the original schematics the
4. Building the Measurement Set-up
20
Table 4.1: Prices of the components in the LNA
Component
SK170 FET
2N5639 FET
Resistors, metal-film, SMD
Capacitors, SMD
Capacitors, electrolytic
Inductor
AD825 operational amplifier
LM833 operational amplifier
DPDT Switch
Connector, BNC
Connector, SMA
PCB, FR4
Total price
Amount
2
1
9
8
2
1
1
1
2
2
1
5 cm x 8 cm
Total price / A
C
1
0.1
0.9
0.4
0.5
1
3
2
6
3
1.5
1
19.4
output of the PLL amplifier was between the used supply voltages (±15 V), so it would
have been necessary to alter the amplifier circuit also in this case. The amplifier used in
this thesis could also have been converted to output different tuning voltages, but due to
time limits it was not done.
Component level modifications were also necessary because not all components were
easily available. For example, the SK170 FETs was replaced by the 2SK369 FETs and the
AD825 operational amplifier was replaced by the 310 operational amplifiers. The final
schematics and layout can be found in Appendix A. Table 4.1 lists the approximate prices
of the used components.
Basically the chosen amplifier circuit is a high-gain low noise amplifier with three
different gain and bandwidth options. The low pass section was modified and first created
as an external version to make it easier to measure the performance of the system. In the
end the low pass section was included in the final version of the PCB board.
It is not very difficult to build a suitable amplifier with low noise properties, even when
using ordinary operational amplifiers, although care must be taken when choosing the
different parts. For example carbon film resistors are not suitable because they have poor
noise properties, but metal film and wirewound resistors are good candidates. Operational amplifiers and FETs have really varying noise properties, depending solely on their
intended use, and this must also be considered when choosing the parts. [10]
Basically this amplifier is a basic voltage amplifier, but it has some features which
make it good for this use. For example the amplifier has three different gain settings
which can be used to avoid overloading the amplifier and blocks following it. Another
way to avoid overloading devices after the amplifier is to use switchable attenuators, but
it is good to have alternatives. Sometimes one method will work better than the other, as
was noticed when performing the actual measurements.
4. Building the Measurement Set-up
21
Switchable attenuator or switchable gain is necessary because of the huge dynamic
range of the measured signal. As already stated, the required dynamic range can exceed 120 dB, and usually the usable dynamic range of a spectrum analyzer is somewhere
around 70 to 90 dB. In this context dynamic range is defined as: "The ratio, expressed
in dB, of the largest to the smallest signals simultaneously present at the input of the
spectrum analyzer that allows measurement of the smaller signal to a given degree of uncertainty." [15] It is thus necessary either to attenuate the carrier or amplify the noise. The
actual measurement procedure is described in more detail in section 5.4.
The amplifier has three different bandwidth/gain-related settings. The 0 dB and 30 dB
gain settings have no additional high pass filter, so it is possible to measure small offset
frequencies. The 60 dB gain setting has an additional high pass filter, a 1 µF capacitor
in series, so that low phase noise far from the carrier can be measured even when strong
noise is present near the carrier. With the 1 µF capacitor the −3 dB frequency should be
around 250 Hz [10]. Here the −3 dB frequency means the frequency where the amplifier
gain has dropped 3 dB from the highest value in the frequency response figure. The actual
performance of the real amplifier is verified in Section 5.2.
In this thesis, the EXT DC modulation mode of the Agilent 8648C is used for the PLL
as described in [5, p. 9.11]. When the signal from the mixer is connected to the MOD
INPUT/OUTPUT connector and the EXT DC modulation is turned on, the frequency
of the output signal of the signal generator is varied relative to the DC voltage at the
connector. Thus the signal generator can work as a part of the PLL circuit, because the
output voltage of the mixer is relative to the phase difference of the two oscillators. When
the signals at the mixer inputs are in quadrature phase, the output is 0 V DC. This voltage
varies to either to positive or negative direction depending solely on the phase difference
between the input signals. The DC voltage is used to tune the frequency of the reference
oscillator to get 0 V at the input of the signal generator.
Figure 4.3 shows the final amplifier construction. SMA was used as the input connector, and BNCs for the outputs. This amplifier is not an RF amplifier, so there is no
real need to use SMA connectors, but the mixer output connector is a SMA and thus no
adapter or special cable is required.
There are two switches in the LNA for the gain changes. The upper one is for choosing
between a buffer and amplification. When the upper switch is down, the amplifier acts as
a buffer, and when it is up, the amplifier amplifies the signal. The lower switch chooses
the gain setting. At the right position there is 30 dB of gain and at the left position there
is 60 dB of gain. Figure 4.4 shows the different switch positions and how they control the
gain settings.
4. Building the Measurement Set-up
22
Figure 4.3: The PCB of the low noise amplifier
(a) Buffer
(b) 30 dB gain
(c) 60 dB gain
Figure 4.4: Switch settings
4.3
Spectrum Analyzer
Care must be taken when deciding which spectrum analyzer to use in this measurement
set-up. Not all RF spectrum analyzers can be used. The biggest problem is that the
frequency range of the spectrum analyzer must be suitable and it may prohibit using the
particular analyzer. Take for example the bandwidths of the two spectrum analyzer which
can be found in the TUT RF Laboratory: Hewlett-Packard 8595E 9 kHz. . . 6.5 GHz and
Agilent E4407B 100 Hz. . . 26.5 GHz. E4407B has two coupling options, ’AC’ with an
internal DC block, and ’DC’ with no DC blocks. With the AC coupling the frequency
range starts at 10 MHz and with the DC coupling it is crucial to remove any DC voltage
before the signal gets to the spectrum analyzer because it does not withstand any DC
voltage in its RF input.
The E4407B spectrum analyzer may be usable if there is nothing else available and
the restrictions have been taken care of, but it would be better to find a spectrum analyzer
4. Building the Measurement Set-up
23
designed for audio frequencies. One additional big problem is that the level at the input of
the spectrum analyzer can become too high if the PLL becomes unlocked and the beat note
appears. The level difference between the carrier frequency power and noise power can
exceed 120 dB even with many VCOs and usually the usable dynamic range of a spectrum
analyzer is much less than that. Attenuators must thus be used when measuring the level
of the carrier and they must be taken off when measuring the noise level. The carrier level
is measured from the beat note, and this is described more clearly in section 5.4.
The amplifier used in this thesis can deliver a signal with approximate amplitude of
9 VRMS to its output when the 60 dB gain setting is used, and the input of E4407B can
tolerate only +30 dBm = 1 W. Even though the output impedance of the amplifier is
not 50 Ω, the input impedance of the spectrum analyzer is and the performance of the
spectrum analyzer will be the limiting part in this system, so 50 Ω is thus used here. The
output impedance of the amplifier is low, so it should be able to deliver its full output
voltage to the 50 Ω load. The maximum output level of the LNA for the 50 Ω load can be
calculated as
U 2 (9 VRMS )2
=
≈ 1.62 W ≈ 32.1 dBm
(4.1)
PAmpOut =
Z0
50 Ω
Agilent does not state the input level which will break the spectrum analyzer, and probably
32.1 dBm would not be absolutely harmful, but it is still better not to exceed the stated
maximum input level +30 dBm. The maximum input level is +30 dBm when all the
internal attenuators are set to their maximum settings. Another thing to consider is that
the level at the mixer inside the spectrum analyzer should be under −10 dBm. If higher
input levels are present, input attenuators inside the spectrum analyzer should be used.
That is one reason why the amplifier used in this system was built with switchable gain
settings. When using the 30 dB gain the output should be under the limits, so with that
setting E4407B could be used without any problems. Still some caution is required during
the measurement process.
When measuring phase noise, the most interesting frequencies are usually between
0 Hz and 100 kHz relative to the carrier frequency, and in the PLL method the carrier is
mixed to 0 Hz, so it is thus the lowest interesting frequency. It is not crucial to have the
bandwidth starting exactly from 0 Hz, but it can be high pass filtered a little higher. In this
set-up the −3 dB corner frequency is around 20 Hz. The highest interesting frequency,
on the other hand, can be little higher in some cases. In E4407B datasheet, phase noise
is shown even up to 10 MHz. The upper −3 dB corner frequency of the amplifier used in
this thesis is about 250 kHz, so that is the practical limit in this case. A more limiting part
may be the spectrum analyzer used. If an audio spectrum analyzer is used, the upper limit
of the bandwidth is probably around 100 kHz. If an RF spectrum analyzer is used, it will
not limit the upper frequency limit.
4. Building the Measurement Set-up
4.3.1
24
About different Spectrum Analyzers
During this project four different spectrum analyzers were tested, and it was found that
the E4407B was the most suitable one, despite the previously stated shortcomings. The
Advantest R9211A was tested first, because it is designed for audio measurements and its
bandwidth (0 to 100 kHz) was suitable. There were some problems with this particular
instrument which made it not suitable for this use. Perhaps the biggest problem was that
it was not possible to get the measured data to a computer for processing. The measured
trace could be printed or saved only in a binary format to a 720k diskette. Because it
was a binary format and there was no program available for reading it, it did not help at
all. Another problem was that this device was really slow for making any measurements,
because it was old and lacked processing power.
Also a basic laptop computer with two different external sound cards was tested. One
of the cards had a 48 kHz sampling frequency so it could reach a little over 20 kHz in
the measurements, which is far from enough in this set-up. This card was used in the
beginning to verify that the built PLL method set-up works at all. Another sound card
had a 96 kHz sampling frequency, so it could reach to over 45 kHz which is still low,
but can be used if nothing else is available. There are also sound cards with a 192 kHz
sampling frequency, which results in adequate bandwidth of about 95 kHz. That kind of
a card was not available, so it was not tested during this project. The measurements with
the USB sound cards were made using the Smaart v.6 as the analyzer [18]. It is designed
for acoustical measurements, but can be used for any raw spectrum measurements up to
the highest frequency available from the analog to digital converter.
4.3.2
Problems in using a Laptop Computer as a Spectrum Analyzer
There are some big problems when using a laptop computer based measurement system,
but they can be solved if the used devices and measurement circuits are designed properly.
Laptops in general exhibit really poor performance in audio measurements. Their power
supplies and internal grounding schemes are not designed for interconnecting to complex
systems. A complex system can be as simple as a laptop and a signal generator separated
by some meters and connected to different power outlets. Usually the internal sound cards
are also not suitable for measurement work.
This noisy sound card problem can usually be disregarded when using external sound
cards which connect to a computer via USB or Firewire. It sometimes is also necessary to
disconnect the power supply of the laptop when making the actual measurement because
when doing that the noise floor of the computer may be lower, and nasty ground loops
may be avoided. This power supply problem was seen once during the project, but it
could not be repeated, so no data is available. Probably the used two laptops were not so
problematic ones.
25
5.
TESTING THE SYSTEM
A system without any performance verification is not very useful. If the performance of
all components is not verified, it is impossible to know whether the measurement results
can be trusted or not. The amount of details which must be gathered and verified from a
specific component varies a lot, but the engineer must know what is required.
In this chapter the components used in the measurement set-up are measured. Their
performance is verified and compared to known standards or published performance specifications if possible. First there is some discussion about measurement basics, followed
by a description of all the preparative measurements where the performance of the different parts was verified, and finally there are the results from the actual phase noise
measurements. For each block separately, it is described which parameters are essential
to measure and why, and what problems they can cause to the final results. At the end of
this section can be found some discussion about the uncertainties in the PLL method.
5.1
Measurement Basics
Phase noise is measured relative to the carrier, as already stated, and thus the difference
between the levels is important, not the absolute signal levels. Signal levels must still be
roughly known, because there are limitations in the measurement equipment. It must be
verified that the signal does not exceed the maximum input level of devices. It should
also be verified that the signal level is not too low, because thermal and instrument related
noise is usually a bigger problem with low signal levels than with high.
First there will be some preparative measurements where all the used equipment is
checked and performance verified. Some parts of the equipment were calibrated by an accredited calibration laboratory in August 2009, so their performance was taken as granted.
After those performance checks, the real measurements will follow.
Table 5.1 shows the oscillators which were measured in this thesis. From now on, their
respective characters will be used for referencing to a particular oscillator. The oscillators
A–C were made as student projects at the Department of Electronics. Oscillators D and E
are Agilent signal generators. Oscillator E offers two different oscillator modes. In Mode
1 the phase noise performance is optimized for offset frequencies under 10 kHz and in
Mode 2 it is optimized for offset frequencies over 10 kHz. These two modes will also be
used in different measurements, and it will be stated which one was used in each case. The
modes will be called E1 and E2, respectively. Figure 5.1 shows the published phase noise
5. Testing the System
26
Table 5.1: Oscillators measured in this thesis
#
A
B
C
D
E
Frequency (MHz)
1800
422
420
0.009-3200
0.25- 3000
Pout (dBm)
−3
−15
8
−10 /−5
−10 /−5
Designer / Model
Student project 1
Student project 2
Student project 3
Agilent 8648C
Agilent ESG-D3000A
Figure 5.1: The published phase noise performance of the ESG-D4000A (Published with permission) [19]
specification of ESG-D4000A. Figure 5.2, in turn, shows the phase noise curves measured
with the PLL method during this project. Even though the published specification is from
a different model, the phase noise properties are probably similar within the same series.
In addition, the published specification only tells the typical performance of this series,
and there can be some differences in the actual measured performance. The actual phase
noise performance is probably better, because it is unwise to publish a specification which
is much better than the actual performance of a device.
5.2
Verification of the Measurement Set-up
When verifying that the measurement setup works as designed, it is necessary to measure
all components individually and then gradually connect them together and verify that
all the components work together as a system. This section describes the verification
process. It is divided in parts based on the amount of measurements and details related
5. Testing the System
27
−80
Oscillator E1 @ 422 MHz
Oscillator E2 @ 422 MHz
Phase Noise Power Density / dBc/Hz
−90
−100
−110
−120
−130
−140
1k
10k
Frequency / Hz
100k
Figure 5.2: Measured phase noise when using different PLL modes of the ESG-D3000A
to the measured components. After the verification process described in this section, the
whole set-up can be built and final measurements made.
5.2.1
Agilent E4407B Spectrum Analyzer
First measurements were made with a Agilent E4407B spectrum analyzer and two signal
generators D & E using the measurement set-up shown in Figure 3.5. The phase noise
module of the spectrum analyzer was compared to the phase noise module of the KE5FX
GPIB Toolkit [16], which was already described in section 3.5. The results seemed to be
too close to each other, with only the 3 GHz trace a little different, as shown in Figure 5.3.
3 GHz is the highest available frequency of ESG-D3000A, so it is not surprising that the
performance is poorer than in the middle of the frequency range. Both the signal generators have too good phase noise properties, to be measured with the spectrum analyzer,
causing the phase and other noise properties of the spectrum analyzer to dominate. The
conclusion is that this method can not be used to measure these oscillators. The phase
noise figure of E4407B can be seen in Figure 5.4 and when comparing that figure to the
measurement results in Figure 5.3, it is easy to see that the spectrum analyzer is the dominant noise contributor. Later, more noisy oscillators were measured directly with the
spectrum analyzer, and the results from these measurements can be found in section 5.3.
5. Testing the System
28
−70
Oscillator D
Oscillator E2 @ 422 MHz
Oscillator E2 @ 3 GHz
Spectrum Analyzer Specification
Phase noise density / dBc/Hz
−80
−90
−100
−110
−120
−130
100
1k
10k
Frequency / Hz
100k
1M
Figure 5.3: Phase noise of different oscillators, measured with the phase noise module of E4407B.
∆-points show the published phase noise specification of E4407B [20].
5.2.2
Mini-Circuits ZX05-42MH Mixer
Figure 5.5 shows the measurement set-up for verifying that the mixer would be suitable.
Agilent 8648C was used as the oscillator. Its output was fed to the power splitter (MiniCircuits ZN2PD-9G) and then through two cables with different lengths to the mixer.
These cables caused different amounts of delay to the signals. There would thus be some
phase difference between the two signals at the mixer inputs resulting in a DC voltage at
the output. The DC voltage was measured with a basic voltmeter. A real phase shifter
would have been of help in this set-up, but it was not available. With such a device the
phase could have been adjusted more gently, but the cable method worked as well for this
basic check. At this point the only interesting part is whether the output would be a DC
signal relative to the phase difference of the input signals or not.
The electrical lengths of the cables were measured with the Agilent 8722D vector
network analyzer and they were 2.7 ns and 8.8 ns. The DC level was 89.43 mV when the
frequency was 1809 MHz. The output power of the signal generator was 0 dBm during
these measurements. 1809 MHz was chosen for the test frequency, because the bandwidth
of the power splitter is 1700 to 9000 MHz and 1809 MHz offered the biggest DC level
near the lower limit of the power splitter.
It is important to have a rough idea of the power levels in different parts of the measurement setup. With 0 dBm = 1 mW output power and 50 Ω system the output voltage
5. Testing the System
29
Figure 5.4: Phase noise specifications of Agilent E4407B (Published with permission) [20]
of the signal generator is:
URMS =
p
p
PZ0 = (1 mW) (50 Ω) ≈ 224 mV
(5.1)
Because the signal goes through the power splitter it is attenuated to −3.4 dBm ≈ 0.457 mW
[21] and thus at the input of the mixer the signal levels are (the cable attenuation is approximated to be zero):
URMS =
p
p
PZ0 = (0.457 mW) (50 Ω) ≈ 151 mV
Delay 1
Reference
oscillator
Power
Splitter
Delay 2
Figure 5.5: The measurement set-up used to measure the mixer DC level
(5.2)
5. Testing the System
30
1
V(out) / Volts
0.8
0.6
0.4
0.2
−180
−135
−90
−45
45
90
135
∆φ / degrees
180
−0.2
−0.4
−0.6
−0.8
−1
Figure 5.6: Normalized mixer output voltage vs. phase difference of inputs
Because of the losses, mostly the conversion loss of the mixer, the DC level at the output of the mixer is lower than the level of the input signals. The conversion loss can be
obtained from the specification sheet of the mixer, but it is not suitable for these calculations, because of the way it is defined: The specification assumes that the level of the LO
signal is much higher than the level of the RF signal and that there is a difference in the
frequencies. In the measurement set-up used in this thesis, the levels should be similar
and the frequencies are the same.
If the output level of the mixer to the 50 Ω load is VRMS = 90 mV as was measured
with the voltmeter (high impedance input), the output power of the mixer is

Pout,mixer,dBm = 10 log10 
U2
Z0
1 mW

 = 10 log10
"
#
(90 mV)2
≈ −7.9 dBm
(50 Ω) (1 mW)
(5.3)
The power level has dropped about 4.5 dB inside the mixer.
Figure 5.6 shows the theoretical output of an ideal mixer with varying phase difference
between the input signals. The figure is normalized, so that the maximum output level
is 1 [22]. The exact DC level is not important in this measurement method, because the
DC is filtered away before the measurements and the interesting part is the ripple in the
DC voltage, which is related to the phase noise of the oscillator. The ripple was measured
with an oscilloscope, and it was from some millivolts to a couple of dozens of millivolts,
so it is a low level signal. The ripple was observed using the same oscillator as a source
and varying its output frequency, so the behavior of the ripple could be observed with the
varying phase difference between the input signals.
5. Testing the System
31
Figure 5.7: The schematics of the Butterworth low pass filter
Reference
Oscillator
DUT
Oscilloscope
Figure 5.8: Measurement set-up used for measuring the (amplitude) response of the low pass filter,
the attenuators and the LNA
5.2.3
Low Pass Filter, Attenuators, Cabling and Adapters
The high frequency content of the signal after the mixer must be removed. The interesting
part of the signal is between 0 Hz and about 1 MHz. With the carrier frequencies in the
gigahertz range, it is not necessary to have a sharp cutoff right after the highest frequency
of interest. It is thus a good idea to build a filter with only a few components and thus
with few non-idealities. The pass band of the filter must also be flat.
With these specifications in mind, it was decided that a Butterworth LC-type low pass
filter would be the right one. Figure 5.7 shows the schematics of the chosen filter. The
filter was built around a 2.2 µH inductor which was available and had a suitable form
factor for mounting on the PCB. Some calculations for the required capacitor values were
made with a program offered by Dale Heatherington [23]. The calculations showed that
with the available inductor and 454 pF capacitors, the −3 dB corner frequency should be
around 7 MHz. The frequency is a little bit high, but it was still decided to build the filter
with these components.
The frequency response of the filter was measured with the set-up shown in Figure 5.8.
The test signal is a sine wave of the required frequency. The first channel of the oscilloscope measures the signal level before the DUT and the second one after the DUT. The
signal frequency is varied inside the chosen frequency range and both the input and output
levels of the DUT are measured. From the results the gain is calculated and converted to
decibels. This same set-up was used for measuring the attenuator settings and the frequency response of the LNA. Because the lowest available frequency of the RF signal
generators in the RF Laboratory was still too high, the Thurlby-Thandar TG230 function
generator with a more suitable frequency range was used.
5. Testing the System
32
1
0
−1
−2
Gain / dB
−3
−3 dB point ↑
−4
@ 4.2 MHz
−5
−6
−7
−8
−9
−10
10k
100k
1M
10M
Frequency / Hz
Figure 5.9: Measured frequency response of the low pass filter, −3 dB point at 4.2 MHz
Table 5.2: Measured performance of the attenuators used
#
A
B
Atten. / dB
0–110
0–110
Bandwidth / GHz
0–18
0–4
−10 dB
−9.9
−10.8
−20 dB
−20.1
−30 dB
−29.3
-
−40 dB
−36.7
-
According to measurements, the filter was suitable for use in the measurement set-up.
As Figure 5.9 shows, the −3 dB corner frequency is lower than was designed, around
4.2 MHz, but it is actually better this way. There are various reasons for this drop in the
corner frequency. The values of the different components are not exact, but these variations alone can not alter the frequency as much as it changed. The filter was designed for
50 Ω source and load impedances, and this is not the situation with the real system. The
source impedance is around 50 Ω because the source is the mixer, but the load impedance
is something totally different. The exact value is not known, but is quite high because of
the voltage amplifier input that follows the low pass filter. In addition, the performance
of the filter was measured with an oscilloscope with an input impedance around 1 MΩ +
X pF.
Some attenuators were also used in the different parts of the measurement system and
their settings verified. Table 5.2 shows the attenuators used and their measured specifications at different attenuator settings. The measurements were made with the same set-up
as shown in Figure 5.8.
There are no specific requirements for the cabling or adapters. All cables and adapters
from the RF Laboratory can be used. Because the set-up utilizes devices and components
5. Testing the System
33
with different connectors, it is important to plan the cabling in a way that minimizes the
number of adapters. The locations of various devices and the routing of cables should also
be considered to minimize errors caused by external electro-magnetic fields and crosstalk.
5.2.4
Low Noise Amplifier
Figure 3.1 shows that the PLL method includes an LNA. It is located after the mixer and
the low pass filter. The amplifier was described in more detail in Section 4.2 and in this
section the amplitude response of the amplifier is verified.
The amplitude response of the amplifier was measured with two different methods.
The first method is shown in Figure 5.8. The second method uses broadband noise as the
test signal and the Agilent 4407B spectrum analyzer as the measurement instrument. The
noise level is measured with the gain of the amplifier set to the different settings (0, 30
and 60 dB). The actual gain and the frequency response can be obtained by calculating
the difference in level between 0 and 30/60 dB gain settings. This method assumes that
the 0 dB setting has no gain or attenuation. The response was measured with the external
high-pass filter, because the high-pass filter affects the frequency response of the amplifier
circuit.
Figure 5.10 shows the results from the amplitude response measurements. The black
dashed lines show the −3 dB thresholds. There are two traces for both gains, because the
frequency response was measured with two different methods. Due to measurement restrictions in the second method, caused by the Agilent E4407B spectrum analyzer, results
from this measurement start at a somewhat higher frequency.
The 0 dB setting was not measured very thoroughly here, because it is only a buffer, and
its frequency response is assumed to exceed the other limiting parts. Quick verifications
confirm this assumption: it works as a 0 dB buffer at least up to 1 MHz.
Figure 5.11 then shows more clearly how flat the response is in the bandwidth of
0.5–100 kHz. This range was chosen for closer investigations because it is the main
range for the phase noise measurements. It is the range where the amplifier has the flattest
response and thus there is no need to make any corrections for the measured phase noise
because of non-flat gain of the amplifier. The trace shows the difference to the mean value
of the gain in the chosen region, at specific frequencies. It can be seen that the gain starts
to decrease when approaching 100 kHz, but because the ripple stays under ±0.5 dB, it
can be neglected.
As Figure 5.10 shows, the frequency response of the amplifier is close to the specifications. The high-pass −3 dB frequency was specified to be around 20 Hz and this is
verified in the 30 dB trace. Also the lower −3 dB frequency of the 60 dB is close to the
specified one, which was 250 Hz. In those frequencies which are covered by both the
oscilloscope and the spectrum analyzer measurement results, the traces are so close to
5. Testing the System
34
65
60
55
Gain / dB
50
60 dB/osc
60 dB/SA
30 dB/osc
30 dB/SA
−3 dB level
Bandwidth
45
40
35
30
25
20
10
100
1k
Frequency / Hz
10k
100k
Figure 5.10: The frequency response of the amplifier. Osc means that it was measured with the
oscilloscope method and SA that it was measured with the spectrum analyzer method. Bandwidth
means the bandwidth which was chosen for closer investigations shown in Figure 5.11.
each other that the results seem reliable. In the 30 dB trace the biggest difference is about
0.3 dB and in the 60 dB trace it is about 0.4 dB.
5.3
Phase Noise Measurements with the Spectrum Analyzer
One of the free running oscillators (B) was stable enough to be measured with the phase
noise measurement utility of E4470B and KE5FX GPIB Toolkit [16]. It was measured
three times with the cabling and other settings staying the same all the time. Figures 5.12
and 5.13 show the results. As the figures clearly show, it is hard to do the measurements
accurately when the measured oscillator is a free running one. There are big variations in
the 0.1–10 kHz region and from 5 kHz up the performance of the spectrum analyzer degrades the accuracy, as can be observed from Figure 5.4. The variations in low frequencies
depend probably on the stability of the carrier frequency during the measurement period.
Sometimes it varies more and sometimes less, and because it is a random process by
nature, no two traces will be similar.
As was already seen in Figure 5.3, the phase noise performance of the spectrum analyzer is not good enough for serious measurements. This spectrum analyzer method is an
easy method to use when the DUT falls in the right performance range so that it can be
measured. At least with the E4407B spectrum analyzer used in this project the range just
is so small that almost no oscillator falls in it. Either the phase noise performance of the
oscillator is too good, or the frequency of the oscillator drifts too much. In both cases the
result is that phase noise cannot be measured with this method.
5. Testing the System
35
0.5
Gain 30 dB
Gain 60 dB
0.4
0.3
0.2
∆Gain / dB
0.1
0
−0.1
−0.2
−0.3
−0.4
−0.5
1k
10k
Frequency / Hz
100k
Figure 5.11: The variation in gain compared to the mean gain in the shown bandwidth
The performance of some oscillators was measured also with the spot-frequency mode
of the phase noise module of the spectrum analyzer. The results are shown in Figure 5.14.
Phase noise was at first measured only up to 100 kHz, because that is the upper limit of
the PLL method. There was an anomaly at 100 kHz so it was decided to extend the range
to 150 kHz to observe the behavior of the method in larger offset frequencies. The spot
frequency mode works a little better than the direct phase noise measurement method of
the E4407B when the DUT is more noisy, even though there is a defect with the measurement module. The module changes resolution bandwidth (RBW) at specific offset
frequencies, and the measurement module does not take this into account. It assumes that
the RBW stays the same all the time. The RBW changes at 30 kHz and 100 kHz offset
frequencies, and at larger offset frequencies a correction factor must be subtracted from
the measurement results. It can be calculated from the following equation [24]
H = 10 log10
RBW
1 Hz
(5.4)
The H values with different RBW values must be calculated, and the the smallest one
subtracted from the other values. This is the correction factor which must be subtracted
from the results in larger offset frequencies. So, the RBWs are 1 kHz, 3 kHz, 10 kHz for
offset frequencies up to 30 kHz, 100 kHz, and 150 kHz respectively. H values are thus
30 dB, 34.77 dB, and 40 dB. With offset frequencies up to 30 kHz nothing is subtracted.
5. Testing the System
36
Oscillator B
Oscillator B
Oscillator B
Spectrum Analyzer Specification
−20
Phase noise density / dBc/Hz
−40
−60
−80
−100
−120
−140
100
1k
10k
Frequency / Hz
100k
1M
Figure 5.12: Phase noise of the oscillator B measured three times with the E4470B
With 30 to 100 kHz, 4.77 dB is subtracted and with 100 to 150 kHz, 10 dB is subtracted
from the results.
Some of the traces in Figure 5.14 start in the larger frequencies, because those oscillators were too noisy, and the measurement could not be locked to the carrier in those
low offset frequencies. When the offset frequency is larger, the accuracy for the carrier
frequency can be lower and that is the reason why the measurement could be locked with
the larger offset frequencies. The accuracy can be lower, because with larger offset frequencies the phase noise changes less with the same frequency change. The level of phase
noise falls more rapidly near the carrier than far from it, as Figure 2.2 shows.
5.4
Actual Phase Noise Measurements using the PLL Method
The schematics of the final set-up can be found in Figure 5.15, and Figure 5.16 shows
a photograph of the final set-up. There is an optional section in the schematics which
can be omitted if the output level of the DUT is in the right region, and if the carrier
frequency does not drift too much. The optional attenuator can be used to bring the level
near −5 dBm at the input of the mixer. With greater levels the output of the mixer would
be distorted. The exact level depends on the mixer used. Too low a signal level would
also be a problem. A low noise amplifier can be used to boost the signal from the DUT,
if necessary. In addition, if the DUT drifts much, it is necessary to somehow measure the
frequency when doing the measurements. It can be done with the optional FFT/spectrum
analyzer.
5. Testing the System
37
Figure 5.13: Phase noise of the oscillator B measured five times with the KE5FX GPIB toolkit
5.4.1
Measurement Procedure
The actual measurement procedure consists of two different parts. The level of the carrier
is measured first. It is done by disconnecting the PLL loop and mis-adjusting the reference
oscillator by a few dozens of kilohertz so that there appears a beat note whose frequency is
visible in the spectrum analyzer. The frequency of the beat note is the difference between
the two oscillators. [22]
In this thesis the frequency of the beat note was usually between 30 kHz and 60 kHz.
Figure 5.17 shows the beat note on the screen of the spectrum analyzer. The level of the
peak (Pcarrier ) is measured.
It is important to use low enough drive levels so that the beat note is undistorted. When
observing the beat note on the display of the spectrum analyzer this can be verified. If the
second and third harmonics of the beat note are more than 30 dB lower than the level of
the beat note, the signal can be called undistorted. If the harmonics are too high level, the
oscilloscope shows a distorted sine wave, so this can also be used as a guide. If the signal
is distorted and drive levels can not be lowered, some corrections can be made. These
corrections are described in [22] and [7].
After the amplitude of the beat note has been measured, the PLL circuit is connected
again and the frequency of the reference oscillator is tuned the required amount to get the
PLL locked. When the PLL is locked, there is a DC level shown on the screen of the
oscilloscope. The input of the oscilloscope must be set to DC coupling and the triggering
5. Testing the System
38
−80
Phase noise density / dBc/Hz
−90
−100
−110
−120
−130
−140
Oscillator C
Oscillator A
Oscillator B
Oscillator E1
Oscillator E2
1k
10k
Frequency / Hz
100k
Figure 5.14: Phase noise measured with the spot frequency mode
also for DC. The frequency should be adjusted in a way to get the DC level as close to
0 V as possible. Figure 5.18 shows the error introduced by a DC level far from 0 V. The
measured oscillator was E and the used frequency was 422 MHz.
When the PLL is locked, it is time to measure the phase noise levels. Figure 5.17 shows
the measured noise level. At this point it is necessary to do some amplitude compensation,
to take care of the limited dynamic range of a spectrum analyzer. It can be done either by
switching more gain to the amplifier, or less attenuation to the attenuator located just before the spectrum analyzer. The latter one was found easier. When more gain is switched
on, PLL usually becomes unlocked, the beat note appears, and the reference oscillator
must be adjusted again. At this point it is important to remember that even with 30 dB of
gain from the LNAE, the beat note may be of too high level for the spectrum analyzer and
thus it is recommended to have some switchable attenuator before the spectrum analyzer.
When measuring noisy oscillators, it was sometimes necessary to keep hands on the attenuator and quickly add some attenuation if the PLL became unlocked. If the amplitude
of the beat note is high enough, it can break the spectrum analyzer even if attenuation
is quickly added. That is why it is important to be very careful during this phase of the
measurement. This unlocking can be seen easily and quickly on the oscilloscope, so the
oscilloscope should stay connected all the time when doing the measurements.
The effect of different gain settings was also tested, but the results were not promising.
It seems to be more convenient to use the same amplifier gain at all times, and just use
the switchable attenuator for the amplitude compensation. This way the PLL stays locked
5. Testing the System
39
FFT/
Spectrum
Analyzer
Mini-Circuits
ZX05-42MH
1
DUT
Power
Splitter
ATT/
LNA
Reference oscillator
HP 8646C
7
2
MOD INPUT/
OUTPUT
Optional section
PLL
Measurement
Signal
6
FFT/
Spectrum
Analyzer
9
3
LNA
5
For checking the
state of the PLL
Oscilloscope
ATT
4
8
Figure 5.15: The schematics of the final measurement set-up. Numbering is the same in Figure 5.16. The switch 9 is actually the Ext Mod On/Off -button in the spectrum analyzer. It is used
to turn the PLL off for the beat note measurement.
better and it is faster and more convenient to do all the measurements. If the amplifier
was located outside of the PLL, the gain changes would not alter the PLL and this method
could be used better. On the other hand, then it would not be possible to change the loop
gain of the PLL and the PLL might not work at all.
5.4.2
Processing the Measurement Results
When the noise level is measured, it has to be processed to get the actual values. The
amplitude compensation must be taken care of, values must be transformed to show noise
power in a 1 Hz bandwidth, and all additional amplitude related corrections have to be
considered. The following equation shows how these different variables affect the result.
All variables are in the decibel scale. [5, p. 9.12], [22]
LSSB (dBc/Hz) = PN − Pbeat − Gcomp − 10 log10 (B) − Gcorrection
(5.5)
where LSSB is the actual phase noise density, PN is the measured noise level in dBm, Pbeat
the measured level of the beat signal in dBm, Gcomp includes all attenuator or amplifier
settings in dB, B is the equivalent noise bandwidth of the spectrum analyzer IF filter in
Hz, and Gcorrection includes all other necessary amplitude corrections in dB. These terms
will be described in detail in the following text.
Probably the most interesting part in Eq. (5.5) is the − log10 (B). Because phase noise
density is expressed as power in a 1 Hz bandwidth and the measurement usually uses a
5. Testing the System
40
Figure 5.16: The final measurement set-up. The beat note is visible on the spectrum analyzer (6)
and on the oscilloscope (8). Numbering is the same in Figure 5.15.
different bandwidth (100 Hz was used in this thesis), the measured level must be transferred to show the actual noise level in the 1 Hz bandwidth. It can be done with the
logarithm shown in Eq. (5.5), where B is the equivalent noise bandwidth of the used filter. This is not directly the resolution bandwidth (RBW) of the spectrum analyzer, or
frequency resolution of the FFT device, but it must be multiplied with a correction factor.
This correction factor depends on the shape of the filter and the result is called equivalent
noise bandwidth. Two different filters may have the same 3 dB bandwidths, but different
equivalent noise bandwidths.
The correction factor depends on the shape of the filter used. If the shape is known, it
can be calculated or found from literature. The shape is not always known well enough.
For example Agilent Technologies states that they use ’Gaussian like’ or ’near-Gaussian’
filters and that the equivalent noise bandwidth is about 1.05 to 1.13 times the 3 dB bandwidth [15]. This depends on the used RBW. The manual of the E4407B spectrum analyzer says that with the RBW in the range of 1 to 300 Hz the filter shape is ’Digital,
approximately Gaussian shape’ and with the RBW in the range of 1 kHz to 5 MHz it
is ’Synchronously tuned four poles, approximately Gaussian shape’. If the correction
factor is omitted, there is a small error in the results. With 100 Hz RBW the differ-
5. Testing the System
41
−20
Beat note
Noise
−26.79 dBm
−30
−40
Signal Power / dBm
−50
−60
−70
−80
−90
−100
−110
10
20
30
40
50
60
Frequency / kHz
70
80
90
100
Figure 5.17: The beat note and measured noise level as seen on the screen of the spectrum analyzer
ence in B is 0.53 dB between no correction factor and 1.13 as the correction factor.
log10 (1 · 100 Hz) = 20 dB and log10 (1.13 · 100 Hz) ≈ 20.53 dB In this thesis a correction factor of 1.1 was used.
In Eq. (5.5) the last part Gcorrection includes all necessary amplitude correction parameters. Because of the measurement method of most spectrum analyzers, based on the
superheterodyne principle, the measured noise level is lower than the actual noise. An
instantaneous amplitude of noise has no meaning, because the amplitude of noise is often
Gaussian distributed, and can theoretically have any value from zero to infinity. A noise
level averaged over some time period is a better way to express the amount of noise. The
so called power averaging, which is proportional to RMS voltage, satisfies this requirement. [15]
The Gaussian distributed noise, with standard deviation σ, is band limited as it travels
through the IF section of the spectrum analyzer, and it becomes Rayleigh distributed.
Video filtering, or averaging is used to get a steady value, the mean value. The Rayleigh
distribution has a mean value of 1.253σ. The analyzer is actually a peak-responding
voltmeter calibrated to indicate the RMS value of a sine wave. Thus the analyzer must
scale its readout by √12 = 0.707 = −3 dB to get RMS value from peak value. Also the
mean value of the Rayleigh distributed noise is scaled with the same amount, and that
√
gives us reading that has a mean value of 1.253σ
≈ 0.886σ (−20 log10 (0.886) ≈ 1.05 dB
2
below σ). This error is constant, and thus the result can be corrected by adding 1.05 dB
to the displayed values. [15]
5. Testing the System
42
−80
DC level far from 0 V
DC level close to 0 V
Phase Noise Power Density / dBc/Hz
−90
−100
−110
−120
−130
−140
1k
10k
Frequency / Hz
100k
Figure 5.18: Difference in the measured phase noise when using the different DC levels
The display scale of the analyzer is also one error source, but fortunately the error is
also a constant and can be corrected. If a logarithmic scale is used, the output of the envelope detector is a skewed Rayleigh distribution. The mean value of this is 1.45 dB lower
than the mean value of a non-skewed Rayleigh distribution. Thus the total correction factor is 2.5 dB. The gain of the used logarithmic amplifier is a function of signal amplitude,
so lower noise values are amplified more than the higher values. [15]
With an Agilent ESA or PSA Series spectrum analyzer the need for the 2.5 dB correction depends on the used averaging method. There are three different averaging methods:
video, voltage, and power. If power averaging is used, no correction is needed because it
determines the average RMS level by the square of the magnitude of the signal, not by the
envelope of the voltage. In this thesis power averaging was used during the measurements,
so the 2.5 dB correction is not needed. [15]
As was stated already, if the phase noise level of the reference oscillator is close to the
phase noise level of the DUT, an additional correction factor must be subtracted from the
result. This was discussed in Section 3.2. In this thesis, a 3 dB correction factor was used
when measuring the oscillator E, and no correction was used for other oscillators. The
reference oscillator in all measurements was the Agilent 8648C signal generator. It was
assumed that the phase noise of the 8648C and the oscillator E (Agilent ESG-D3000A
signal generator) would be similar and thus the correction factor must be used. It was also
assumed that the free running oscillators A-C would have worse phase noise performance.
5. Testing the System
43
−80
−5 dBm
−10 dBm @ day 2
−10 dBm @ day 1
Phase Noise Power Density / dBc/Hz
−90
−100
−110
−120
−130
−140
1k
10k
Frequency / Hz
100k
Figure 5.19: Repeatability of the measurement
When mixing signals down to DC, one sideband of the signal is folded onto the other.
Because the noise voltages are in phase, they add coherently, and thus additional 6 dB
must be subtracted from the measured noise power density. [22] Now the correction
factor for the oscillator E can be calculated: Gcorrection = 3 dB+6 dB = 9 dB. For other
oscillators Gcorrection = 6 dB.
Figure 5.19 show how well this measurement with the PLL method can be repeated
if all the measurement settings and cables used are the same. The same oscillator (E2,
frequency 1 GHz) was measured on two different days and naturally the measurement
set-up was disassembled and built again in between. On the first day the measurement
was repeated five times (traces −10 dBm @ day 1) and the frequency of the reference
oscillator was adjusted to get the PLL out of lock between all measurements. The biggest
variation is located somewhere around 6 kHz and it is about 2 dB as can be seen from
the figure. On the second day it is similar with about 3 dB difference. This difference is
probably caused by changes in cabling and adapters, and supply voltage to the LNA. The
supply voltage level was not calibrated between the two measurement sessions, but it was
set approximately to ±15 V.
When the output levels of the oscillators were set to −5 dBm, the output is surprisingly
different, as can be seen in the third trace group (−5 dBm) in Figure 5.19. This is interesting, because with the higher drive levels the mixer behaved more linearly, if we use
the the criterion discussed in the start of this section. With a −10 dBm oscillator power
the harmonics were only 29.5 dB lower but with −5 dBm the difference was about 38 dB.
5. Testing the System
44
−55
Oscillator B
Oscillator C
Oscillator A
−60
Phase Noise Power Density / dBc/Hz
−65
−70
−75
−80
−85
−90
−95
−100
−105
−110
1k
10k
Frequency / Hz
100k
Figure 5.20: Phase noise of the oscillators A, B and C
According to these measurement results, the measured phase noise performance is about
13 dB poorer when using higher drive levels.
Despite this behavior, it can be said that this measurement is reliable and time invariant,
when measuring good enough oscillators with the same set-up.
5.4.3
Measuring the free running Oscillators
Finally there are some results from the other oscillators measured during this project. The
results can be seen in Figure 5.20. Although it is always interesting to measure devices
which are not commercial off-the-shelf products, it was decided to not use those when
verifying the performance of the measurement set-up. When verifying something, it is
essential to have as few variables as just possible. In this case the amount of variables
was made lower by using a good enough oscillator as a DUT. When it was verified that
the set-up would work as planned, some real life measurements were made. This kind of
oscillators will be measured with this system in the future, so it is a good idea to have also
some understanding about that process. It should be noted that the scale of the verticalaxis has been changed, so when comparing these traces to the other traces this should be
considered.
It was noticed that even with this kind of a measurement setup, noisy oscillators are
problematic. They can be measured, but the results have much variation, and thus it may
be necessary to use more statistical processing. The same oscillator should be measured
many times, and some statistical methods used to calculate a average phase noise curve.
5. Testing the System
45
The statistical methods do not usually take systematic errors in the account, so those
must be handled separately. Statistical methods are not covered in this thesis, and it is
recommend that the reader should familiarize himself with them if reliable results are
wanted.
5.5
Error Analysis
The basis for getting the final phase noise density results with the PLL method (Equation (5.5)) is repeated here for convenience
LSSB (dBc/Hz) = PN − Pbeat − Gcomp − 10 log10 (B) − Gcorrection
The exact error limits in all terms are unknown, but this section will give some insight in
the amount of deviation in the results that can be expected.
It was noticed during the measurements that even when all the settings should be similar, there can be 3 dB difference between the measured noise levels. So, it can be assumed
that the variation in PN is ±1.5 dB. There are some areas in the measurement set-up which
cause this variation. First of all, even if the same cables and adapter are used all the time,
there are some variation in how well the connectors are mated and how cables are routed.
Also a small difference in the supply voltage of the low noise amplifier can cause some
variation, as already discussed. The temperature around the measurement set-up will also
have an effect.
The variation in Pbeat (the measured beat note power) depends mostly on the spectrum
analyzer, and can be assumed small, around ±0.25 dB. The accuracy of Gcomp (the amplitude compensation) depends on the method used to measure all the attenuators and LNA.
±0.25 dB is a realistic assumption for this term, because the measured attenuation values
were compared to the calibrated values and they were inside the ±0.2 dB range.
The biggest variation in 10 log10 (B)-term is caused by the shape factor of the applied
filter. Some discussion can be found at Section 5.4.2. B is calculated as B = RBW ·
S. With the resolution bandwidth (RBW) set to 100 Hz, the difference in −10 log10 (B)
term between the shape factor S = 1 and S = 1.13 is 0.53 dB. So, a realistic variation is
±0.25 dB.
For the last term, Gcorrection (includes all additional corrections), a realistic assumption
for the variation is ±0.5 dB. The variation in the last term is caused by problems in determining how close the phase noise performances of the DUT and the reference oscillator
are to each other.
Experimentally it was seen that a ±2 dB variation in phase noise density is a realistic
expectation for the final results. This can be seen in Figure 5.19. Some more in depth
error analysis can be made with the root-sum-square (RSS) uncertainty or the maximum
uncertainty methods. Even though these methods are not designed for handling numbers
5. Testing the System
46
Table 5.3: The uncertainty results
Term
Value
PN
−80. . . −60 dBm
Pbeat
20. . . 40 dB
Gcomp
−30. . . −20 dB
10 log10 (B)
20. . . 21 dB
Gcorrection
6. . . 11.5 dB
RSS uncertainty
Maximum uncertainty
Uncertainty
±1.5
±0.25
±0.25
±0.25
±0.5
±1.6 dB
±2.8 dB
in the decibel scale, it is a practice in the RF field to use them for that. Using numbers in
the decibel scale introduces some error, but with small numbers the error is also small, so
it can be neglected.
Reference [25, p. 19] defines the RSS error as
s
E=
n
∑
i=1
∂f
E
∂Xi X̄i
2
(5.6)
where f is the function under investigations, here the sum, Xi is a parameter of f , and EX̄i
is the estimated error in term Xi . The maximum error occurs when the maximum error is
realized in all the terms and all the errors are in the same direction [25, p. 19]:
n
∂f EX̄
Emax = ∑ i,max
∂X
i
i=1
(5.7)
So, the RSS error estimate for LSSB is
q
2
2
EP2N + EP2beat + EG2 comp + E10
log10 (B) + EGcorr
p
√
= 1.52 + 0.252 + 0.252 + 0.252 + 0.52 = 2.6875 ≈ 1.64 (dB)
ELSSB =
(5.8)
The ±1.64 dB error is quite close to the ±2 dB error that was estimated experimentally,
so it is quite safe assumption. The maximum error in this measurement is:
ELSSB ,max = EPN + EPbeat + EGcomp + E10 log10 (B) + EGcorr
= 1.5 + 0.25 + 0.25 + 0.25 + 0.5 = 2.75 (dB)
(5.9)
Table 5.3 gathers together the uncertainty results with the information about the order of
magnitude of all the terms.
47
6.
CONCLUSIONS
In this chapter the whole thesis is summarized. Conclusions are drawn from the different
parts and the thesis in general. The goals of the thesis are discussed and the results covered. There are also some suggestions what can be done to make the system better, and
what things could not be covered in this thesis but should be investigated and verified in
the future.
6.1
Goals of the Thesis
The goals of this thesis were to investigate phase noise and different measurement systems, and to build and verify the performance of one phase noise measurement system
that could be used in the RF laboratory. These goals were fulfilled, although some things
were left open for future research.
A usable measurement system resulted from this project. It is still an experiment version without any casing, for example. It was verified that the system could be used for
phase noise measurements, and that was one of the goals. It can thus be said that the
project was successful.
6.2
Future Research
The noise properties of the low noise amplifier were not verified. Experimentally it was
seen a low noise one when using the 30 dB gain settings, but no measurement data was
gathered. Its performance should thus be verified and probably some fine tunings could
be made to make it even less noisy.
The build quality and form factor of the amplifier could be improved. At the end of this
project it is just a plain PCB board without any case. A metal case with real connectors,
especially for the supply voltage, should be considered. It would also be wise to include to
the board the now external DC block, which protects the spectrum analyzer, because then
two connectors could be excluded from the system, and probably the noise properties
would also improve a little. In this context build quality means that the manufactured
amplifier was just a test version built to verify the performance of this method. It can be
used for actual measurements, but it is not made to withstand the usual student laboratory
handling.
It must also be said that the current PCB layout has some problems. It was probably the
tenth version, and was altered dramatically just before the last version which was used for
6. Conclusions
48
Signal
Generator
DUT
Low
Pass
FFT/
Spectrum
Analyzer
DC
Block
Oscilloscope
(a) Set-up used in this thesis
Signal
Generator
DUT
Low
Pass
FFT/
Spectrum
Analyzer
DC
Block
Oscilloscope
(b) Alternative set-up
Figure 6.1: Two variations of the same measurement set-up
the actual measurements. There is for example not enough space for the supply voltage
connectors. To reduce the required amount of adapters, the other output BNC connector
should be changed to an SMA connector, and it would probably be wise to include a third
output connector.
An even better improvement would be to include also the mixer to the system. Right
now it was time consuming to build and dismantle the measurement setup. It took usually
about 15 to 20 minutes to assemble the system, because many adapters and different cables had to be used. A better system would have inputs from the DUT and the reference
oscillator and outputs for the spectrum analyzer, PLL, and the oscilloscope. If the connectors are chosen properly there will be no need for adapters. With careful planning and
design even the switchable attenuator could be included in this device. It would make the
measurement process much easier and results more repeatable, because fewer connections
would have to be made.
One other interesting variation is to get signal to the PLL and oscilloscope right after
the mixer and low pass filter. The difference is shown in Figure 6.1. The amplifier would
amplify only the signal going to the spectrum analyzer. By doing it this way, the amplitude
compensation during the measurement process could be more easily made with the gain
switch of the amplifier. The down side is that then it would be impossible to tune the loop
gain of the PLL. It should naturally be verified, whether this would work at all, because
the loop gain could not be changed.
6. Conclusions
49
It would be also interesting to study whether a noisy VCO could be measured better
with the set-up used in this thesis, or with the alternative setup where the reference oscillator is the free running one, and the DUT is included in the PLL loop. This could not
be tested during this process, because a suitable PLL amplifier was not available and the
tight schedule did not allow to build one. A suitable PLL amplifier takes the output of the
mixer (± some dozens of millivolts) and generates a suitable tuning voltage for the VCO.
The suitable tuning voltage is usually 0 to 5 V DC, but a good PLL amplifier would have
an adjustable tuning voltage.
Even though there are some topics left for future research, it can be said that the measurement system built during this project can be used to measure phase noise of oscillators. Measuring phase noise of low noise oscillators is not an easy task, but it can be
done without expensive commercial phase noise measurement systems using only the
equipment available in most RF laboratories.
50
REFERENCES
[1] Peter Vizmuller. RF Design Guide: Systems, Circuits and Equations. Artech House,
1995.
[2] Mini-Circuits. VCO Phase Noise. [WWW] [Cited 2010-04-28] Available at: http:
//www.minicircuits.com/pages/pdfs/vco15-6.pdf.
[3] Hewlett Packard. RF & Microwave Phase Noise Measurement Seminar. [WWW]
[Cited 2010-04-28] Available at: http://www.hparchive.com/seminar_notes/
HP_PN_seminar.pdf.
[4] Ray Kammer. A Speech at the SIMNet Launch. [WWW] [Cited 2010-04-27] Available at: http://www.nist.gov/speeches/rk-simnet.html.
[5] ARRL. The ARRL Handbook 2010. ARRL – The National Association for Amateur
Radio, 2010.
[6] Stephan R. Kurz. WJ Tech note – Mixers as Phase Detectors. [WWW] [Cited 201005-11] Available at: http://www.triquint.com/prodserv/tech_info/docs/
WJ_classics/Mixers_phase_detectors.pdf.
[7] Agilent Technologies. Phase Noise Characterization of Microwave Oscillators
– Phase Detector Method – Product Note 11729B-1. [WWW] [Cited 2010-0510] Available at: http://www.home.agilent.com/upload/cmc_upload/All/
5952-8286E.pdf.
[8] Warren F. Walls. Practical Problems Involving Phase Noise Measurements.
[WWW] [Cited 2010-04-28] Available at: http://tycho.usno.navy.mil/ptti/
ptti2001/paper42.pdf.
[9] Aeroflex. Application Note #2 –PN9000 Automated Phase Noise Measurement System. [WWW] [Cited 2010-04-28] Available at: http://www.datasheetarchive.
com/datasheet-pdf/010/DSA00173368.html.
[10] Wenzel Associaties.
Techniques for Measuring Phase Noise.
[WWW]
[Cited 2010-04-28] Available at:
http://www.wenzel.com/documents/
measuringphasenoise.htm.
[11] Charles Wenzel. A Low Noise Amplifier for Phase Noise Measurements. [WWW]
[Cited 2010-04-28] Available at: http://www.wenzel.com/pdffiles1/pdfs/
lowamp.pdf.
REFERENCES
51
[12] Radio-Electronics.com. Coax Cable Velocity Factor. [WWW] [Cited 2010-0427] Available at: http://www.radio-electronics.com/info/antennas/coax/
coax_velocity_factor.php.
[13] Wenzel Associates. Products of Wenzel Associates, Inc. [WWW] [Cited 2010-0428] Available at: http://www.wenzel.com/products1.htm.
[14] Cristoph Rauscher. Fundamentals of Spectrum Analysis. Rohde & Schwartz GmbH
& Co, 2007.
[15] Agilent Technologies. Agilent Spectrum Analysis Basics – Application Note 150.
[WWW] [Cited 2010-04-28] Available at: http://cp.literature.agilent.
com/litweb/pdf/5952-0292.pdf.
[16] John Miles. User-Friendly Phase-Noise Measurement for HP, Tektronix and Advantest Spectrum Analyzers. [WWW] [Cited 2010-04-28] Available at: http:
//www.thegleam.com/ke5fx/gpib/pn.htm.
[17] AML Communications. Techniques for Measuring Ultra-Low Phase Noise in Microwave Amplifiers. [WWW] [Cited 2010-04-28] Available at: http://amlj.com/
files/products\%20page/phasenoise-technique.pdf.
[18] Rational Acoustics. Sound system measurement, optimization and control software
for Microsoft Windows and Mac OsX. [WWW] [Cited 2010-03-14] Available at:
http://www.rationalacoustics.com/pages/Smaart_Landing_Page.
[19] Agilent Technologies. Agilent PSA and ESA-E Series Spectrum Analyzers Phase
Noise Guide (Option 226). Agilent Technologies, 2001.
[20] Hewlett-Packard. User’s Guide – HP ESG-D Series Signal Generators. HewlettPackard, 1997.
[21] Mini-Circuits. The Data Sheet of ZN2PD-9G+ Power Splitter/Combiner. [WWW]
[Cited 2010-05-19] Available at: http://minicircuits.com/pdfs/ZN2PD-9G+
.pdf.
[22] Morris Smith. Freescale Semiconductor AN1639 – Phase Noise Measurement Using
the Phase Lock Technique. [WWW] [Cited 2010-04-28] Available at: http://
cache.freescale.com/files/rf_if/doc/app_note/AN1639.pdf.
[23] Dale Heatherington. Design an L-C Low pass or High pass Filter. [WWW] [Cited
2010-04-28] Available at: http://www.wa4dsy.net/filter/hp_lp_filter.
html.
[24] Agilent Technologies, Markku Salonen. Private Communication. 2010-04-19.
REFERENCES
52
[25] Lehto, A. & Räisänen, A. Mikroaaltomittaustekniikka. Otatieto, 2001.
[26] CadSoft Computer GmbH. EAGLE Layout Editor. [WWW] [Cited 2010-05-12]
Available at: http://www.cadsoft.de/.
53
A.
LNA SCHEMATICS AND PCB LAYOUT
This appendix includes the schematics and the PCB layout of the low noise amplifier
designed for the PLL method measurement set-up. The sizes of the etching masks are
correct, so they can be directly copied if the reader wants to build the same amplifier and has the same parts available. The PCB was designed with Eagle Layout Editor [26], and the Eagle files of this amplifier can be obtained by sending an email to
olli.rajala@ravoltek.net.
A. LNA Schematics and PCB Layout
54
## 4K
k.M
4MMM
.
7
4K
k
X
4X
M94
##34K
!0.
C
4MM
4Q
4X
## 4K
CX
k
C .
4MM
K
K
.4
4MM
CXM
S
kKk
.M
K
C
.
KC7Q
!04
!04
SKM
4S
7
##34K
4K79.
C 4
K
k
##34K
.
K
4M5
4
Q
S
..
C
X77
.7
4
7
749C
.
4
4
M94
4Q
.
k
.
.
C
4
7
.M
M94
k
.
7
S
.
7
..M
##34K
4
..M
34K
4K
## 4K
##34K
## 4K
Figure A.1: The schematics of the low noise amplifier
4C
M94
X77
47
X
77M
k
.
.4SMk.
X.M5
!0.
4
4
.4SMk.
.
A. LNA Schematics and PCB Layout
55
Table A.1: LNA parts
Part
C1
C2
C3
C4
C5
C6
C7
C8
C9
C13
C18
C19
C20
IC2
IC4
Q1
Q2
Q6
R15
R16
R17
R18
R19
R20
R21
R22
R23
S1
S2
U$1
X1
X2
X3
Value
220u
220u
454p
1000p
100p
820 p
68u
454p
10 u
0.1u
0.1u
0.1u
0.1u
LM833D
AD825
2K17042N
2K17042N
2N5639
420
330
153.2
750
680
5
100
1k
31.6k
Omron A9T
Omron A9T
2.2u
Package
SMC_D
SMC_D
C1206
C1206
C0805K
C1206
E/7260-38R
C1206
C1206
C1206
C1206
C1206
C1206
SO08
SO08
TO92TO92TO92R1206
R1206
R1206
R1206
R1206
R1206
R1206
R1206
R1206
A9T21-0011
A9T21-0011
own inductor
BU-SMA-V
A1944
A1944
A. LNA Schematics and PCB Layout
m ott o B
56
A. LNA Schematics and PCB Layout
57
Top